A Delphi-method-based consensus guideline for definition of treatment-resistant depression for clinical trials. — TEI-tagged view

Luca SforzinipersName, Courtney WorrellpersName, Melisa KosepersName, Ian M. AndersonpersName, Bruno AouizeratepersName, Volker AroltpersName, Michael BauerpersName, Bernhard T. BaunepersName, Pierre BlierpersName, Anthony J. ClearepersName, Philip J. CowenpersName, Timothy G. DinanpersName, Andrea FagiolinipersName, I. Nicol FerrierpersName, Ulrich HegerlpersName, Andrew D. KrystalpersName, Marion LeboyerpersName, R. Hamish McAllister-WilliamspersName, Roger S. McIntyrepersName, Andreas Meyer-LindenbergpersName, Andrew H. MillerpersName, Charles B. NemeroffpersName, Claus NormannpersName, David NuttpersName, Stefano PallantipersName, Luca PanipersName, Brenda W. J. H. PenninxpersName, Alan F. SchatzbergpersName, Richard C. SheltonpersName, Lakshmi N. YathampersName, Allan H. YoungpersName, Roland ZahnpersName, Georgios AislaitnerpersName, Florence Butlen-DucuingpersName, Christine FletcherpersName, Marion HaberkamppersName, Thomas LaughrenpersName, Fanni-Laura MäntyläpersName, Koen SchruerspersName, Andrew ThomsonpersName, Gara Arteaga-HenríquezpersName, Francesco BenedettipersName, Lucinda Cash-GibsonpersName, Woo Ri ChaepersName, Heidi De SmedtpersName, Stefan M. GoldpersName, Witte J. G. HoogendijkpersName, Valeria Jordán MondragónpersName, Eduard MaronpersName, Jadwiga MartynowiczpersName, Elisa MellonipersName, Christian OttepersName, Gabriela Perez-FuentespersName, Sara PolettipersName, Mark E. SchmidtpersName, Edwin van de KetterijpersName, Katherine WoopersName, Yanina FlossbachpersName, J. Antoni Ramos-QuirogapersName, Adam J. SavitzpersName, Carmine M. PariantepersName (2021date). “A Delphi-method-based consensus guideline for definition of treatment-resistant depression for clinical trials..” Molecular Psychiatrytitle·j.

Object of study — the definition of treatment-resistant and partially responsive depressionterm·dom for clinical trialsterm·key

Research question — how TRDterm·dom and PRDterm·dom should be defined and measured for future clinical studies

Contribution — operational definitions of TRDterm·dom and PRDterm·dom with graded consensus recommendations for trial design

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

A Delphi-method-based consensus guideline for definition of treatment-resistant depressionterm·dom for clinical trialsterm·key.

Abstract Criteria for treatment-resistant depressionterm·dom (TRDterm·dom) and partially responsive depressionterm·dom (PRDterm·dom) as subtypes of major depressive disorderterm·dom (MDDterm·dom) are not unequivocally defined. In the present document we used a Delphi-method-based consensus approach to define TRDterm·dom and PRDterm·dom and to serve as operational criteria for future clinical studies, especially if conducted for regulatory purposes. We reviewed the literature and brought together a group of international experts (including clinicians, academics, researchers, employees of pharmaceutical companies, regulatory bodies representatives, and one person with lived experience) to evaluate the state-of-the-art and main controversies regarding the current classification. We then provided recommendations on how to design clinical trialsterm·key, and on how to guide research in unmet needs and knowledge gaps. This report will feed into one of the main objectives of the EUropean Patient-cEntric clinicAl tRialterm·key pLatforms, Innovative Medicines Initiative (EU-PEARL, IMI) MDDterm·dom project, to design a protocol for platform trials of new medications for TRDterm·dom/PRDterm·dom.

Introduction Many medications have proven efficacy in major depressive disorderterm·dom (MDDterm·dom) [ 1 , 2 ], but frequently and even with multiple medication exposures, they fail to improve MDDterm·dom symptoms [ 3 – 7 ]; one third of individuals do not achieve full symptomatic remissionterm·key [ 3 ], and even fewer meet criteria for both symptomatic and functional remissionterm·key [ 8 ]. In individuals with ineffective initial treatments, even if subsequent treatments are effective, there is a very high relapse rate while continuing the treatment; for example, in the STAR*D trial, individuals who required more treatment steps had higher relapse rates (up to 71% after the fourth step) [ 3 ].

‘Incomplete response’ is not an all-or-nothing phenomenon, but rather a continuum that ranges from partially responsive depressionterm·dom (PRDterm·dom), to treatment-resistant depressionterm·dom (TRDterm·dom), to ‘multi-therapy-resistant MDDterm·dom (MTR-MDD)’ [ 9 ], to ‘refractory depression’, which implies an absence of response to all currently available treatments. Unfortunately, there is a lack of consensus definitions around concepts such as PRDterm·dom, TRDterm·dom, and ‘adequate’ treatments [ 10 – 13 ]. Moreover, guidelines on how to treat TRDterm·dom/PRDterm·dom, such as pharmacological augmentation, are not consistent [ 14 ], and evidence on effectiveness is sparse [ 15 ]. This complicates the generalizability of results from research settings to the real-world, and hinders progress in this field, as there is no uniform population for clinical and biological investigations in TRDterm·dom/PRDterm·dom, including clinical trialsterm·key for new or repurposed medications. Importantly, regulators acknowledge that response, partial response, and non-response exist on a continuum without universally accepted definitions, but nevertheless distinguish between these conditions; indeed, treatments for TRDterm·dom and PRDterm·dom already have accepted regulatory paths for drug approval.

Interestingly, a recent consensus statement (including experts also participating in the present report) suggested that the terms PRDterm·dom and TRDterm·dom are semantically and operationally not ideal, and proposed the broader concepts of ‘difficult-to-treat depressionterm·dom (DTDterm·dom)’ or ‘suspected DTDterm·dom’ [ 7 ], described as “depression that continues to cause significant burden despite usual treatment efforts”. This concept overlaps with PRDterm·dom and TRDterm·dom, but introduces a more flexible, multidimensional and longitudinal definition. The authors themselves acknowledged that “what constitutes significant burden” is “subjective and likely to vary between patients”, thus implicitly involving the patient’s point of view, and also likely to vary among clinicians and raters. Indeed, the authors cautioned that this definition may not be specific and objective enough to define clinical populations for regulatory clinical trialsterm·key. In a subsequent follow-up paper, they also emphasised this as a ‘model of care’ that might be useful for both individuals with MDDterm·dom and clinicians, especially in averting the development of ‘therapeutic nihilism’ [ 16 ].

In this document, we use a Delphi-method-based consensus approach to define TRDterm·dom and PRDterm·dom and to deliver operational criteria for future clinical studies, including clinical trialsterm·key conducted for regulatory purposes. We have reviewed the relevant literature and brought together a group of international experts (including clinicians, academics, researchers, employees of pharmaceutical companies, regulatory bodies representatives, and one person with lived experience (PWLE)) to discuss the current state-of-the-art and the main controversies regarding the current classification. Our specific aims are to: (1) deliver TRDterm·dom/PRDterm·dom definitions that could be used at a person-centred level using information that is currently routinely collected in clinical practice; (2) recommend measures and instruments that should be included in future investigations and clinical trialsterm·key for TRDterm·dom/PRDterm·dom; and (3) indicate which are the important areas that require further research. We provide consensus recommendations and the level of agreement for each recommendation, and discuss their limitations. We balance the need of clinicians and scientists with those of regulatory authorities, and describe differences when indicated.

This initiative is part of the EU Patient-cEntric clinicAl tRialterm·key pLatforms (EU-PEARL) programme, a public-private strategic partnership funded by the Innovative Medicines Initiative (IMI) to conceptualize and lead the design of an integrated research platform (IRP), that is, an infrastructure which allows the planning and completion of platform trials (see Appendix for further information).

Our Delphi-method-based approach The Delphi approach is a method of choice for developing guidelines in health research [ 17 ]. Indeed, a recent commentary to the aforementioned paper on DTDterm·dom [ 7 ] specifically recommends using a Delphi approach for consensus statement in depression [ 18 ]. However, this method could be difficult to standardize, with poor reporting of findings [ 19 , 20 ]. Probably the most important of these issues has been identified in the definition of consensus [ 21 ]. Therefore, we clearly defined consensus as a percentage of agreement (vs. disagreement) on a precise recommendation, and further defined strong consensus when this percentage was equal or above 95%, moderate consensus between 61 and 94%, and weak consensus between 51 and 60%. Details of our Delphi process are available in the Appendix and summarized in Fig. 1 . Briefly, it consisted of: (1) identification of experts with a track record of publications in this area or stakeholders with clear expertise, from clinical practice, academia, industry, and regulatory agencies; (2) first draft report, with an up-to-date narrative review on TRDterm·dom/PRDterm·dom definitions (in February 2020) and a questionnaire to gather opinions on the debated issues; (3) an online consensus meeting with academics and clinicians, on the 22nd of May 2020; (4) second draft report, integrating comments to the first draft with those from the meeting, and again circulated to all contributors; (5) third draft report (with an updated narrative review, in September 2020), circulated to a different group of stakeholders, including representatives of regulatory authorities, industry, and one PWLE; (6) an online consensus meeting with stakeholders, on the 9th of October 2020, with additional stakeholders providing written feedback; (7) fourth draft report (with a systematic review of the literature published from March 2020 to January 2021), integrating stakeholders comments, which was then circulated and approved by all authors, and submitted in its entirety (approximately 28,000 words) internally to IMI EU-PEARL (available from the corresponding author on request); (8) editing of the report into this shorter version for publication, again circulated and approved by all authors. Comments to the different versions of the report were always anonymized. Methodological details of the narrative reviews and of the subsequent systematic review are presented in the Appendix. Fig. 1 Delphi process flow diagram. Schematic representation of the different phases of the Delphi-method-based process.

We provide a number of recommendations, each based on the view supported by the largest number of the experts, while highlighting areas of uncertainty and defining the ‘level of consensus’ on each recommendation: ‘strong’, indicating the top one-third of recommendations that had unanimous or almost unanimous (≥95%) agreement; ‘moderate’, indicating almost all the remaining two-thirds of recommendations, which had a substantial majority (61–94%); and ‘weak’, indicating the one recommendation that only reached the minimum consensus of 51%. We have summarised all the recommendations in Table 1 and Fig. 2 . We suggest that: (1) recommendations with strong consensus should be immediately adopted; (2) recommendations with a moderate consensus could provide a working model for immediate implementation, but should be further confirmed by future research/discussion; and (3) recommendations with a weak consensus represent an area of uncertainty and thus require further research/discussion. Table 1 Main consensus recommendations on TRDterm·dom/PRDterm·dom regulatory clinical trialsterm·key. Level of consensus– Strong % of agreement Recommendations which can be implemented within current practice TRDterm·dom and PRDterm·dom definitions 1 A definition of TRDterm·dom for clinical trialsterm·key conducted for regulatory purposes is necessary. 98% Previous antidepressant treatments 3 TRDterm·dom should be defined after a minimum of two failed treatments with <25% of improvement with adequate dosing and duration. 96% Type of medications 10 Discontinuation of treatment before the completion of the fourth week, without clear evidence of lack of responseterm, should not be considered as a treatment failure for the purpose of establishing TRDterm·dom/PRDterm·dom. 96% Exclusion from TRDterm·dom/PRDterm·dom studies 12 A previous structured psychotherapy failing to improve MDDterm·dom symptoms is not an exclusion criterion. 100% Clinical presentation 15 All specifiers of depression (melancholic, atypical, anxious, psychotic, mixed) should be considered within the TRDterm·dom/PRDterm·dom definition, except for bipolar depression. 95% Recommendations which can be implemented in future research 21 Future research should recognize and target different clinical phenotypes of TRDterm·dom (and PRDterm·dom) underpinned by a specific biological mechanism. 100% 22 For future research, diagnostic and history-taking instruments should be implemented in clinical cohorts and electronic health records, to allow a reliable, comprehensive, and multidimensional evaluation of people with lived experience (PWLE). 100% 24 Preferences, perspectives, and reported outcomes of PWLE should be included in future TRDterm·dom (and PRDterm·dom) diagnostic tools and outcome measures. 100% Level of consensus – Moderate Recommendations which can be implemented within current practice TRDterm·dom and PRDterm·dom definitions 2 It is important to distinguish between TRDterm·dom and PRDterm·dom for randomized clinical trialsterm·key for new treatments. 85% Previous antidepressant treatments 4 PRDterm·dom can be defined even after a single treatment (improvement 25 to <50%) with adequate dosing and duration. 76% 6 It is possible to assess ineffective past/current antidepressant treatment attempts, but only if properly documented, that is, based not only on subjective recollection or standardised instruments to assess psychiatric history and previous treatments (see below), but also on clinical documentation. 75% Type of medications 7 To define TRDterm·dom, the two antidepressant treatment failures should consist of two established (licensed) medications for MDDterm·dom of different mechanisms of action. 75% 8 A failed course of psychotherapy should not be included as one of the previous treatments required for the definition of TRDterm·dom/PRDterm·dom. 78% 9 The criteria of ‘adequate dose and duration’ is the minimal effective dosage, that is, the minimal approved dosage, administered for at least four weeks. 74% Exclusion from TRDterm·dom/PRDterm·dom studies 11 Multiple-drug resistant individuals, and individuals in whom augmentation strategies failed to improve/eliminate MDDterm·dom symptoms should not be excluded from TRDterm·dom/PRDterm·dom studies. 93% 13 Individuals with MDDterm·dom in whom deep brain stimulation (DBS) and vagus nerve stimulation (VNS) failed to improve/eliminate MDDterm·dom symptoms should be excluded from TRDterm·dom/PRDterm·dom clinical studies. 62% 14 Individuals with MDDterm·dom in whom other non-continuous/non-invasive brain stimulation interventions, such as electroconvulsive therapy (ECT) or transcranial magnetic stimulation (TMS), failed to improve/eliminate MDDterm·dom symptoms should not be excluded. 88% Clinical presentation 16 Comorbid personality disorders or other mental disorders should be excluded from TRDterm·dom/PRDterm·dom studies only when their onset is properly documented as independent and antecedent to the MDDterm·dom diagnosis. 79% 17 Individuals with a severe substance use disorder not currently in remissionterm·key should be excluded from TRDterm·dom/PRDterm·dom studies, independently from the onset; in contrast, individuals with comorbid substance use disorder that is active and mild/moderate should be excluded from TRDterm·dom/PRDterm·dom studies only when the onset is properly documented as independent and antecedent to the MDDterm·dom diagnosis. 83% Diagnostic tools and measures of outcome 18 Maudsley Staging Modelterm·key is the preferred instrument to assess TRDterm·dom/PRDterm·dom status. 69% 19 Clinician administered MADRS10 is the preferred outcome instrument to assess treatment responseterm (and remissionterm·key), together with patient-reported QIDS-SR. MADRSterm·key = 85% QIDS-SR = 81% 20 Criteria for remissionterm·key, response, and partial response should not be relaxed in regulatory clinical trialsterm·key for TRDterm·dom/PRDterm·dom, and shorter versions of the traditional scales, such as the HAM-D6 and the MADRS6, should not be preferred to full scales. 92% Recommendations which can be implemented in future research 23 Currently, no biomarker has been validated in clinical practice or in clinical trialsterm·key to identify people with TRDterm·dom (and PRDterm·dom), or to further stratify them; however, collection of biological samples for subsequent subgroup or stratified analyses is recommended. 91% 25 The usefulness of adherenceterm·key assessment using blood levels or other methods (also in a run-in period) should be assessed through research, before deciding whether it should be implemented in future clinical trialsterm·key. 92% Level of consensus – Weak Recommendations which can be implemented within current practice Previous antidepressant treatments 5 The definition of TRDterm·dom should include two treatment failures both within the current episode, and the definition of PRDterm·dom should include partial response to at least one treatment within the current episode; moreover, for long current episodes, only treatment failures within the last two years should be considered. 51% Fig. 2 Algorithm for inclusion in TRDterm·dom/PRDterm·dom studies. Recommended algorithm for inclusion of participants in TRDterm·dom and PRDterm·dom studies based on the consensus recommendations discussed throughout the manuscript and listed in Table 1 .

TRDterm·dom and PRDterm·dom definitions What is lack of responseterm?

A treatment responseterm in MDDterm·dom is defined by a reduction of at least 50% in MDDterm·dom severity [ 22 ] on a standardised rating scale (see below); ‘lack of responseterm’, encompassing both TRDterm·dom and PRDterm·dom, is the mirror image of this definition, that is, less than 50% reduction in depression severity. TRDterm·dom is further associated, in most definitions, with the ‘classic criteria’ of lack of responseterm to at least two medications at an adequate dose and duration (see below). There is, however, much variability around the definition of TRDterm·dom. In a recent systematic review, Brown et al. [ 12 ] reported that, out of 155 TRDterm·dom definitions identified in the published literature, 48.4% specified at least two sequential treatment failures as a requirement. In an even more recent systematic review, Gaynes and colleagues [ 5 ] found that only 37% of intervention studies in TRDterm·dom had enrolled individuals with MDDterm·dom meeting the criteria of at least two failed antidepressants, and only 19% had also described failure to adequate doses and durations of treatments. In fact, the most common definition for TRDterm·dom in intervention trials involved a minimum of only one previous failed treatment (48%). Thus, the majority of studies on TRDterm·dom do not seem to use the ‘classic criteria’, making it difficult to pool or compare data across these TRDterm·dom studies. Not surprisingly, there was a strong consensus (98%) for our first recommendation that a definition of TRDterm·dom for clinical trialsterm·key conducted for regulatory purposes is necessary .

Operational criteria for TRDterm·dom and PRDterm·dom Based on a number of guidelines and other expert documents [ 22 – 25 ], in this report we adopt the definition of TRDterm·dom as indicating individuals who show a reduction of less than 25% in MDDterm·dom severity to at least two antidepressants, and of PRDterm·dom as indicating individuals who show a reduction of between 25% and <50% in MDDterm·dom severity to at least one antidepressant. This should ideally be established using prospective psychometric assessments, or at least using clinical interviews and health records to measure retrospectively the improvement in depression severity (see below).

The majority of experts, even though with moderate consensus (85%), recommended the importance of this distinction between TRDterm·dom and PRDterm·dom for randomized clinical trialsterm·key for new treatments, especially because of the potential advantage of separating these individuals for different types of randomized controlled trials (for example, switching for TRDterm·dom vs. augmentation for PRDterm·dom).

Previous antidepressant treatments Here we discuss: (1) how many previous treatments should be considered, (2) in which episode (current and/or past) and (3) the way the treatment failure or partial response should be assessed for inclusion in clinical trialsterm·key (prospectively and/or retrospectively).

Number of previous treatments The FDA guidance [ 26 ] acknowledges that no universally accepted definitions exist for TRDterm·dom or PRDterm·dom, and proposes that TRDterm·dom studies should include people with MDDterm·dom who have not responded to more than one (so, at least two) prior medications administered at an adequate dose and duration. Similarly, the EMA opts for the same definition of at least two failed treatments when considering the matter “in a clinical pragmatic view”. This same definition is confirmed as the most common in the review by Gaynes and colleagues [ 5 ], and endorsed by the DTDterm·dom consensus statement by McAllister–WilliamspersName and colleagues [ 7 ], although the last document states that some individuals might be considered to have DTDterm·dom even with a single treatment failure, for example when standard treatments are contraindicated. There is also an uncertainty around PRDterm·dom. EMA documents do not specify a precise number of previous treatments to diagnose PRDterm·dom [ 27 ], while the FDA indicates that “for adjunctive treatment, studies should include patients with partial responses to other antidepressant therapies” [ 26 ].

Consistent with the most commonly used criteria, we recommend that TRDterm·dom should be defined after a minimum of two failed treatments with <25% of improvement with adequate dosing and duration (strong consensus, 96%), while PRDterm·dom can be defined even after a single treatment (improvement 25 to <50%) with adequate dosing and duration (moderate consensus, 76%).

Current or past episodes The preferred definition of TRDterm·dom for clinical trialsterm·key includes a current failure and a past failure, i.e., subjects are currently receiving an antidepressant and they are still depressed according to current clinician’s assessment, and they were also treated with another antidepressant in the past and showed no response based on retrospective assessment. However, it is unclear whether both of these treatments should apply to the same (current) episode or to clearly distinct episodes. Indeed, although the EMA definition of MDDterm·dom [ 27 ] emphasizes the current episode for the characterisation of the disease, it does not clarify whether the two failures should both be during the current episode. Of course, it is difficult to retrospectively defining the response to an antidepressant, especially if the current episode is of long duration (years) and the previous treatment was closer to the onset of the episode.

We recommend (with weak consensus, 51%) that the definition of TRDterm·dom should include two treatment failures both within the current episode, and the definition of PRDterm·dom should include partial response to at least one treatment within the current episode; moreover, for long current episodes, only treatment failures within the last two years should be considered. Of note, the “minority” position was split across a continuum of different opinions ranging from those who wanted to maintain that both antidepressants should be in the same episode but within a shorter period of time (shorter than two years) to those who wanted to consider two antidepressants “life-time”; thus, our recommendation sits in the middle of these extremes.

Prospective or retrospective assessment A related issue, particularly important for regulatory clinical trialsterm·key, is whether one treatment failure should be ‘prospective within the trial’, i.e., the trial starts with an established medication for MDDterm·dom at an adequate dose, and then the person is offered a new intervention only if the medication fails to improve or eliminate MDDterm·dom symptoms. We acknowledge that this, while ideal, would lead to operational execution challenges within a trial, with increased complexity and burden for the sites and study participants [ 22 ].

We recommend (with moderate consensus, 75%) that it is possible to assess ineffective past/current antidepressant treatment attempts, but only if properly documented, that is, based not only on subjective recollection or standardised instruments to assess psychiatric history and previous treatments (see below), but also on clinical documentation, such as pharmacy, hospital, or other health records. This documentation can also be used to confirm some degree of adherenceterm·key to the failed treatments, and to screen people with depressive symptoms for previous episodes of mania, hypomania, or sub-threshold bipolarity, since these individuals should be excluded (see below).

Adherenceterm·key to treatment is a well-recognized critical issue in both clinical and research settings. Rates of adherenceterm·key vary across the literature and generally are limited by different and/or restricted time periods [ 28 ]. However, people with MDDterm·dom have typically high reported rates of non-adherence (up to more than 50%) [ 29 , 30 ]. It is therefore vital to properly confirm the individual’s adherenceterm·key in order to define non-response: many cases of TRDterm·dom may not be true TRDterm·dom, but, instead, represent partial or full non-adherence. However, the assessment of treatment adherenceterm·key can be difficult and is often not addressed in everyday settings, or in most current studies in TRDterm·dom. The most reliable method to assess adherenceterm·key is to perform a blood test to measure the concentration of the medication in the plasma (which would also allow the recognition of fast and slow metabolisers), even though it may increase the participants’ burden during the trial. Although this suggestion was clearly supported by the experts, systematic use of plasma level monitoring in TRDterm·dom/PRDterm·dom definitions is not current practice. It is important to note that the FDA does not accept data analyses that exclude individuals not compliant to the previous treatment based on a blood dosing of the medicine. Further discussion on adherenceterm·key is presented in the Appendix. However, we recommend assessing the usefulness of different methods to measure a person’s adherenceterm·key for a potential (and desirable) future implementation (see Future directions).

Type of medications Different classes and mechanisms of action The EMA [ 27 ] mentions that the two treatment failures could be with medications of “same or different class”. Other guidelines, including the FDA guidance [ 26 ], do not make this distinction. Interestingly, the concept of ‘different mechanisms’ may overlap with ‘different classes’, although the pharmacological overlap between classes and mechanism of action is not absolute; the Neuroscience-based Nomenclature (NbN) for psychotropic agents has greatly contributed to clarify this issue [ 31 ]. Moreover, there are some geographically-relevant label limitations to the usage of some medications (see also Appendix).

We recommend (with moderate consensus, 75%) that, to define TRDterm·dom, the two antidepressant treatment failures should be with two established (licensed) medications for MDDterm·dom of different mechanisms of action. The two treatments could be separated by a drug free period (and there is failure to both), prescribed sequentially (switching, with failure to both), or one is prescribed as augmentation to the other (because of the first drug failing, and with the augmentation also failing).

Moreover, we also recommend (with moderate consensus, 78%) that a failed course of psychotherapy should not be included as one of the previous treatments required for the definition of TRDterm·dom/PRDterm·dom, but this information should be reported for staging (see below). Notably, this recommendation does not have implications for the role of psychotherapies in the treatment of TRDterm·dom, and, similarly to other TRDterm·dom models proposed [ 11 ], we support the use of psychotherapies as a treatment strategy for TRDterm·dom in clinical practice.

Dosage and duration of antidepressants Most studies and meta-analyses have found no benefit for antidepressant dose escalation versus staying on the minimum licensed dose, with an increased risk of side effects and discontinuation [ 32 , 33 ], even if there is also evidence that higher doses may have superior efficacy [ 34 , 35 ] and that higher starting doses may be associated with higher response rates [ 36 ]. In terms of ‘minimum duration’, some studies have shown that, while two weeks can be enough to observe an initial response, a stable response/remissionterm·key is usually detectable after four weeks [ 37 , 38 ]. However, a complete remissionterm·key may not be detectable after four weeks. For example, approximately half of remitters on citalopram in level 1 of the STAR*D trial remitted between week 6 and week 14 [ 39 ]. Of note, this is different for fast-acting medications, such as esketamine [ 40 ], but these drugs are indicated for individuals that already have TRDterm·dom.

We recommend (with moderate consensus, 74%) that the criteria of ‘adequate dose and duration’ is the minimal effective dosage, that is, the minimal approved dosage, administered for at least four weeks.

Interestingly, in regulatory clinical trialsterm·key, participants’ discontinuation of treatments may be considered as an endpoint equivalent to a failure [ 41 ]. However, we recommend (with strong consensus, 96%) that discontinuation of treatment before the completion of the fourth week, without clear evidence of lack of responseterm, should not be considered as a treatment failure for the purpose of establishing TRDterm·dom/PRDterm·dom, especially because of the difficulty in distinguishing retrospectively between non-response and intolerance.

Exclusion from TRDterm·dom/PRDterm·dom studies Many people with lived experience (PWLE) who fulfil the proposed criteria for TRDterm·dom may have tried many ineffective antidepressants (multi-drug resistance) or other (licensed or unlicensed) pharmacological augmentation interventions (see also Appendix). We discussed whether these individuals should be considered to have TRDterm·dom which is ‘too severe’ for inclusion in clinical trialsterm·key.

We recommend (with moderate consensus, 93%) that multiple-drug resistant individuals, and individuals in whom augmentation strategies failed to improve/eliminate MDDterm·dom symptoms should not be excluded from TRDterm·dom/PRDterm·dom studies, provided the other criteria are met. We also recommend (with strong consensus, 100%) that a previous structured psychotherapy failing to improve MDDterm·dom symptoms is not an exclusion criterion. Rather, all this information should be recorded for staging and for potential subgroup analyses.

Deep brain stimulation (DBS) and vagus nerve stimulation (VNS) are continuous treatments in which a neurostimulator (usually implanted in the chest wall) is connected to intracerebral electrodes or the left vagus nerve [ 42 , 43 ]. Because of the invasiveness of these intervention, people selected for these treatments have tried an unusually large number of ineffective treatments, and thus are different from other people living with TRDterm·dom [ 44 , 45 ].

We recommend (with moderate consensus, 62%) that individuals with MDDterm·dom in whom DBS and VNS failed to improve/eliminate MDDterm·dom symptoms should be excluded from TRDterm·dom/PRDterm·dom clinical studies. However, we also recommend (with moderate consensus, 88%) that individuals with MDDterm·dom in whom other non-continuous/non-invasive brain stimulation interventions, such as electroconvulsive therapy (ECT) or transcranial magnetic stimulation (TMS), failed to improve/eliminate MDDterm·dom symptoms should not be excluded.

Clinical presentation TRDterm·dom and PRDterm·dom symptoms People living with MDDterm·dom have a range of symptoms specifiers (for example, with melancholic features, with atypical features, with psychotic features), as well as the presence of comorbid diagnoses (for example, comorbid anxiety, bipolar depression). Potentially, some of these symptoms may be more difficult to treat compared with others, such as MDDterm·dom with anxious distress [ 46 – 49 ].

We recommend (with strong consensus, 95%) that all specifiers of depression (melancholic, atypical, anxious, psychotic, mixed) should be considered within the TRDterm·dom/PRDterm·dom definition, except for bipolar depression, which should be excluded as this is part of bipolar disorder.

Comorbidities Individuals with a diagnosis of personality disorder (especially borderline personality disorder) frequently meet criteria for MDDterm·dom, but antidepressants are unlikely to be effective and hence may mimic a non-response. Active substance users also present an increased risk of pharmacological interactions (both pharmacokinetic and pharmacodynamic), side effects, and mood symptoms due to the effects of the substance or of the withdrawal which may appear as a ‘phenocopy’ of MDDterm·dom. However, it is important to highlight that these conditions are so frequently comorbid with depression that they cannot be routinely excluded, otherwise the proposed TRDterm·dom/PRDterm·dom definitions would not be generalisable to individuals seen in everyday clinics.

The FDA guidance also captures this tension between homogeneity and generalisability, stating that “investigators should seek demographically broad populations and avoid unnecessary restriction of study populations (e.g., by excluding patients with concomitant illness)” [ 26 ]. Indeed, the FDA does not explicitly take position on the inclusion or exclusion of personality disorders, and it explicitly encourages to consider people with a history of substance abuse, “although such inclusions should be weighed against concerns about diagnostic and medication effect confounders”, and further states that “patients whose substance use disorder is not at least in partial remissionterm·key will likely be excluded from antidepressant trials depending on the level of particular confounding concerns”. On the other hand, the EMA document only broadly indicates that MDDterm·dom occurring comorbid with other psychiatric disorders is not in the remit of the guideline [ 27 ].

We recommend (with moderate consensus, 79%) that comorbid personality disorders or other mental disorders should be excluded from TRDterm·dom/PRDterm·dom studies only when their onset is properly documented as independent and antecedent to the MDDterm·dom diagnosis. Moreover, in accordance with the FDA guidance [ 26 ], we recommend (with moderate consensus, 83%) that individuals with a severe substance use disorder not currently in remissionterm·key should be excluded from TRDterm·dom/PRDterm·dom studies, independently from the onset; in contrast, individuals with comorbid substance use disorder that is active and mild/moderate should be excluded from TRDterm·dom/PRDterm·dom studies only when the onset is properly documented as independent and antecedent to the MDDterm·dom diagnosis.

Also, somatic comorbidities should be systematically recorded, but not excluded a priori; this includes conditions such as inflammatory, neuroendocrine, and metabolic diseases, which can influence the response to treatments [ 50 – 52 ].

Diagnostic tools and measures of outcome We also established a consensus on the best psychometric tools to measure antidepressant response (and thus TRDterm·dom/PRDterm·dom status and staging) both retrospectively, to improve the way we diagnose TRDterm·dom (and PRDterm·dom) before entering a trial, and prospectively, as tools to use in regulatory trials. A detailed summary of the different psychometric instruments is beyond this core document and is instead presented in the Appendix.

Historical assessment of TRDterm·dom/PRDterm·dom status At a minimum, assessment to define TRDterm·dom should include a structured clinical interview for the diagnosis of MDDterm·dom, such as the Structured Clinical Interview for DSM (SCID) [ 53 , 54 ] and the Mini-International Neuropsychiatric Interview (MINI) [ 55 ], together with a scale to assess the patient’s antidepressant history, such as the Massachusetts General Hospital Antidepressant Treatment Responseterm Questionnaire (ATRQ) [ 56 ], and the Antidepressant Treatment History Form (ATHF) [ 57 ]. A more structured method is to use staging models [ 6 , 58 ], such as the Thase and Rush method [ 59 ], the Massachusetts General Hospital Staging model (MGH-s) [ 22 ], and the more recent Maudsley Staging Modelterm·key (MSM) [ 60 ], which also allows the assessment of previous antidepressant treatment failure, using the Maudsley Treatment Inventory (MTI) [ 10 ].

Among the experts, there was a substantial agreement that staging models are the preferred instrument to define TRDterm·dom/PRDterm·dom and assess the level of treatment-resistance, even if many experts recognize the validity of the ATRQ, often accepted in regulatory clinical trialsterm·key.

We recommend (with moderate consensus, 69%) the Maudsley Staging Modelterm·key as the preferred instrument . However, both the Thase and Rush and the MGH-s models are valid alternatives.

Assessment of depressive symptoms and response to antidepressant treatment The presence and severity of depression are established through clinician-administered scales, such as the Hamilton Depression Rating Scale (HAM-D 17, 21, 24 items) [ 61 ], the Montgomery-Åsberg Depression Rating Scale (MADRS10) [ 62 ], and the Quick Inventory of Depressive Symptomatology (QIDS) Clinician Rating (QIDS-C) [ 63 ]; or self-reported instruments, such as the Beck Depression Inventory (BDI) [ 64 ], the Patient Health Questionnaire-9 Item (PHQ-9) [ 65 ], and the QIDS Self-Report (QIDS-SR) [ 63 ]. When possible, these scales are administered before and after a specific treatment has started, but they can also be used as a single measure to determine the current severity of the depression. Although TRDterm·dom/PRDterm·dom definitions centre on ‘response’, the ideal treatment goal is remissionterm·key, considered as the absence of a relevant MDDterm·dom symptomatology as indicated by a score of ≤7 at the HAM-D17 [ 66 ] or ≤10 at the MADRS10 [ 67 , 68 ].

Of note are also the Clinical Global Impression (CGI) scales, measuring psychiatric global status [ 69 ], and the Sheehan Disability Scale (SDS), assessing functional impairment [ 70 ], which are important since even responders may continue to have significant residual symptoms and functional impairment. Indeed, people with lived experience of depression often define treatment success with an emphasis on broadly-based functional outcomes rather than remissionterm·key of individual symptoms, and their perspective should be included through patient-reported outcomes (PROs) such as quality of life, wellbeing, impact of symptoms, and social and occupational functioning [ 71 ].

We recommend (with moderate consensus) that the clinician administered MADRS10 is the preferred outcome instrument to assess treatment responseterm (and remissionterm·key) (85%), together with patient-reported QIDS-SR (81%). We additionally recommend (with moderate consensus, 92%) that criteria for remissionterm·key, response, and partial response should not be relaxed in regulatory clinical trialsterm·key for TRDterm·dom/PRDterm·dom and shorter versions of the traditional scales, such as the HAM-D6 and the MADRS6, should not be preferred to full scales.

The MADRS10 and QIDS-SR were chosen based solely on experts’ feedback, and not a priori selected (see also Appendix). The MADRS10 was originally developed with the precise aim to have a scale that was more sensitive to change compared with the HAM-D17, a crucial feature for clinical trialsterm·key [ 72 ]. In addition, unlike the HAM-D [ 73 ], it predominately focuses on the core symptoms of depression such as sadness, tension, lassitude, pessimistic thoughts, and suicidal thoughts. This results in a higher internal consistency and a greater accuracy for the MADRSterm·key compared with HAM-D [ 74 ]. However, this is also a potential limitation of the MADRSterm·key, as it may not capture the clinical complexity of the PWLE. Most notably, individuals with MDDterm·dom with non-melancholic or atypical features may present with a range of symptoms that are not captured by the MADRSterm·key. For such reason, there is a risk that MDDterm·dom individuals continuing to report depressive symptoms such as irritability and anxiety may be classified as responders. The same issue may arise with a ‘short’ scale such as the QIDS-SR, which again carries the risk of omitting the measure of MDDterm·dom symptoms that are not captured by standard diagnostic criteria of MDDterm·dom, in contrast to far more granular self-report measures of depressive symptoms, such as the Inventory of Depressive Symptomatology (IDS)-SR [ 75 ], and the Symptoms of Depression Questionnaire (SDQ) [ 76 ]. Thus, the assessment of changes in MDDterm·dom severity (upon which the TRDterm·dom/PRDterm·dom definitions rely) should include the entirety of symptoms that an individual displays, including different (non-melancholic) MDDterm·dom specifiers and symptoms not included in standard classification systems. When possible, the broadest assessment possible should be conducted, using broad-spectrum scales or checklists assessing a wide range of MDDterm·dom related symptoms, such as the HAM-D28. Also, specific instruments assessing specific symptoms, such as anhedonia, motor retardation, or anxiety (for example, the Hamilton Anxiety Rating Scale (HAM-A) [ 77 ]), may be useful. Sub-analyses on specific clusters of symptoms (such as suicidal, atypical, or psychotic symptoms) could be used to generate hypotheses for future clinical trialsterm·key, for medications targeted to specific symptoms or to symptoms-based subgroups of individuals. Another issue is that traditional scales address depressive symptoms in the emotional and physical clusters, with a lower attention on the cognitive ones, which may frequently remain as residual symptoms indicating poor response and thus an increased risk for relapse. Interestingly, there is evidence that a mismatch between scores in clinician-administered and self-reported scales is a poor prognostic sign [ 78 ], supporting the position endorsed in this report to have at least one score of both types of scales included, until new and more tailored instruments will be developed (see also next section and Appendix).

Future directions This section makes recommendations that are not immediately applicable to clinical trialsterm·key and highlights research gaps and new opportunities identified throughout the consultative process. Future trial sponsors may wish to incorporate one or more of these areas to expand the knowledge base. A more detailed discussion of these points is presented in the Appendix; here, we want to emphasise the main recommendations: Future research should recognize and target different clinical phenotypes of TRDterm·dom (and PRDterm·dom) underpinned by a specific biological mechanism (strong, 100%). For future research, diagnostic and history-taking instruments should be implemented in clinical cohorts and electronic health records, to allow a reliable, comprehensive, and multidimensional evaluation of the PWLE (strong, 100%). Currently, no biomarker has been validated in clinical practice or in clinical trialsterm·key to identify subjects with TRDterm·dom (and PRDterm·dom), or to further stratify them; however, collection of biological samples for subsequent subgroup or stratified analyses is recommended (moderate, 91%). Preferences, perspectives, and reported outcomes of PWLE should be included in future TRDterm·dom (and PRDterm·dom) diagnostic tools and outcome measures (strong, 100%). The usefulness of adherenceterm·key assessment using blood levels or other methods (also in a run-in period) should be assessed through research, before deciding whether it should be implemented in future clinical trialsterm·key (moderate, 92%).

Limitations and conclusions While we discuss many points of debate in the Appendix, here we want to acknowledge the main limitations of this report.

First, the recommendations are based on experts’ and stakeholders’ clinical and professional judgement, supported by research and clinical evidence, but without any hard and objective validation. Second, our recommendations try to strike a balance between being ‘too broad’ and being ‘too narrow’, with the aim to identify ‘clinically homogenous’ samples, based on a pragmatic and non-aetiologically-based approach; still, we cannot exclude that significant phenotypic or biological heterogeneity is present within the defined TRDterm·dom/PRDterm·dom samples, and future research will allow to define this inherent variability, if any, in clinical features, biomarkers, and clinical response. Third, the opposite risk is also present: that we define groups so narrowly that the findings are not generalisable to the larger population of PWLE in clinical settings, and do not translate into better care; however, maintaining narrow definitions may help avoiding overmedicalisation and may prevent the lowering of the threshold above which treatments which are specific for TRDterm·dom are offered to patients, as a recent commentary on this topic highlighted [ 18 ].

In terms of theoretical limitations, some of our contributors have argued that the proposed TRDterm·dom/PRDterm·dom definitions are unhelpful from a clinical and a conceptual perspective, as they arbitrarily apply thresholds on a continuum, and are influenced by the different healthcare systems. Moreover, TRDterm·dom and PRDterm·dom concepts are based upon a conceptualisation of depression as an episodic illness with good inter-episode recovery, while many individuals with MDDterm·dom have a chronic illness with waxing and waning course; this might create difficulties when trying to accurately identify the beginning and end of the current/previous episodes, in order to define response to the antidepressants. Finally, we recognise that, according to our recommendations, individuals with MDDterm·dom with an ineffective (<25%) single antidepressant trial are not defined as either TRDterm·dom or PRDterm·dom; as in standard clinical practice, these subjects should be prescribed a second antidepressant before a decision can be made on their TRDterm·dom status.

Notwithstanding these limitations (and the points of uncertainties further discussed in the Appendix), this report offers clear and consistent definitions of TRDterm·dom/PRDterm·dom for regulatory clinical trialsterm·key and for clinical and biological studies more broadly, agreed among a large group of experts including clinicians, academics, industry, regulatory agencies, and PWLE. Our ultimate ambition is to advance tailored treatments and a truly ‘precision medicine’ approach for MDDterm·dom, which in turn will finally help to deliver better care for people suffering from this severely challenging illness, which remains too often ineffectively treated.

Disclaimer This document reflects the majority views of the experts. Authorship reflects having been part of the process and accepting the consensus statements without necessarily personally endorsing every single recommendation (see level of consensus). Also, the views expressed in this article are the personal views of the author(s) and may not be understood or quoted as being made on behalf of or reflecting the position of the regulatory agency/agencies or organisations with which the author(s) is/are employed/affiliated. This publication reflects only the authors’ view, and the JU is not responsible for any use that may be made of the information it contains.

Supplementary information Appendix