Global definition and clinical presentation

Consensus was reached on an age cutoff of ≥65 years for defining TRLLD. The age of 65 years is a conventional geriatric threshold and is also consistent with literature describing clinically relevant differences in depression in older adults. Meta-analytic evidence suggests that antidepressant response may be less robust in adults aged ≥65 years than in younger adult populations, supporting the need for age-specific consideration of treatment resistance from this age onward [Reference Tedeschini, Levkovitz, Iovieno, Ameral, Nelson and Papakostas26]. In addition, LLD is frequently characterised by greater medical comorbidity, vascular burden, neuroendocrine dysregulation, cognitive impairment, and increased risk of subsequent dementia, all of which may influence both clinical presentation and treatment response [Reference Kuo, Lin and Lane27]. Therefore, the ≥65-year cutoff should be understood as a consensus-based and clinically pragmatic threshold [Reference Kuo, Lin and Lane27]. No consensus was achieved on establishing a distinct “very-late-onset” subgroup beginning at age 85. There is an age-related increase in the incidence of new-onset depression across older age strata; however, incidence estimates specifically for individuals aged ≥85 years are less consistently reported [Reference Pálsson, Östling and Skoog28, Reference Weyerer, Eifflaender-Gorfer, Wiese, Luppa, Pentzek and Bickel29]. The depressive phenotype and response to treatment in this oldest group may differ from that observed in the broader older-adult population [Reference Morichi, Dell’Aquila, Trotta, Belluigi, Lattanzio and Cherubini30, Reference Roose, Sackeim, Krishnan, Pollock, Alexopoulos and Lavretsky31]. Older individuals often present with somatic complaints, irritability, neurovegetative changes, and cognitive symptoms, rather than overt dysphoria [Reference Devita, De Salvo, Ravelli, De Rui, Coin and Sergi32]. This aligns with reviews describing ‘masked’ or atypical features, which can hinder the timely identification of TRLLD [Reference Devita, De Salvo, Ravelli, De Rui, Coin and Sergi32]. Further studies should clarify whether incident cases in the oldest-old have distinct features, particularly regarding treatment response, that might justify considering them as a separate category.

Within the panel, the MADRS was the most used monitoring instrument in clinical practice, followed by the GDS or HAM-D/HDRS. This pattern highlights a discrepancy: although atypical presentation was considered important, commonly used symptom scales do not fully capture atypical or masked manifestations. Consequently, these features were not included as core definitional criteria. This may lead to under-detection in routine clinical practice and requires further evaluation in future studies.

Most panellists agreed that non-response to two adequate antidepressant trials should be required, consistent with definitions applied in general adult populations [Reference McIntyre, Alsuwaidan, Baune, Berk, Demyttenaere and Goldberg33]. As summarised in Table 8, although this Delphi study did not define specific target antidepressant doses or trial durations, adequacy should be judged according to recognised national and international guidelines, taking into account both dose and duration [34, Reference Houix, Lepetit, Arbus, Armand-Branger, Bonin-Guillaume and Calvet35]. In line with the principle of “start low, go slow, but go” [Reference Walaszek36], antidepressant treatment in older adults should be initiated and titrated cautiously, while aiming, when clinically appropriate and tolerated, for the same minimum effective doses as in younger adults unless contraindicated [Reference Kok and Reynolds37].

Consensus supported including older adults who develop late-life resistance to an antidepressant that was previously effective. Longitudinal data indicate heterogeneous trajectories in relation to treatment resistance: some patients show inadequate response from the first trial with an antidepressant, whereas others respond initially and then lose response over time [Reference Demjaha, Lappin, Stahl, Patel, MacCabe and Howes38]. These have been termed primary versus secondary treatment resistance, sometimes misinterpreted as early-onset and late-onset [Reference Howes, McCutcheon, Agid, de Bartolomeis, van Beveren and Birnbaum39]. Distinguishing these trajectories is clinically salient in late life: secondary resistance may be more common in those with greater medical comorbidity and may be modifiable [Reference Howes, Thase and Pillinger40]. Additionally, there was strong consensus on distinguishing recurrent versus non-recurrent depression and early- versus late-onset forms – defined by the majority of the expert panel as depression that first manifests at age 65 years or older – given their potentially distinct aetiologies – and, consequently, different treatment implications [Reference Rapp, Dahlman, Sano, Grossman, Haroutunian and Gorman41, Reference Alexopoulos42].

No consensus was reached on whether the mere presence of residual symptoms after two or three adequate treatment trials should independently qualify an individual as having TRLLD. Nonetheless, it is important to emphasise that minimising residual symptoms remains a central therapeutic goal. In later life, these symptoms are common and should be actively evaluated and managed, while also recognising that some may reflect underlying physical illness, frailty, or other non-depressive processes requiring targeted assessment and treatment [Reference Oude Voshaar43]. Where residual symptoms are attributable to ongoing depressive pathology, they are strongly associated with increased relapse risk and poorer quality of life [Reference Dombrovski, Mulsant, Houck, Mazumdar, Lenze and Andreescu44, Reference Hiranyatheb, Nakawiro, Wongpakaran, Wongpakaran, Bookkamana and Pinyopornpanish45].

Experts endorsed the need for a categorical approach to TRLLD and agreed that the Thase and Rush (1997) [Reference Thase and Rush18] staging model should be adapted beginning at Stage II. Stage I – defined as failure of a single adequate antidepressant trial – did not reach consensus, whereas Stage II corresponds to the minimum threshold of the categorical TRLLD definition established in this study (Table 8).

The adapted staging model should be interpreted as a framework for describing the degree of treatment resistance in TRLLD rather than as a prescriptive treatment algorithm. The design of the survey questions addressing this model was informed by the findings of our preceding systematic review [Reference Pozuelo Moyano, Bautista, Ibarra, Mueller, von Gunten and Vandel14]. The differences from the original Thase and Rush model reflect the responses of the Delphi panel across the 2nd SR (see Table 4).

One expert cautioned that the label treatment-resistant can convey a defeatist approach if applied too early in care [Reference Demyttenaere and Van Duppen46], which is especially important given the high non-response rates in LLD, approximately 51% after one antidepressant trial [Reference Gutsmiedl, Krause, Bighelli, Schneider-Thoma and Leucht6]. This concern is further contextualised by meta-analytic evidence concluding that, in adults aged ≥65 years, the overall benefit of second-generation antidepressants (e.g., SSRIs, SNRIs, or mirtazapine) is modest and heterogeneous, and potentially attenuated in the context of frailty and tolerability issues [Reference Tedeschini, Levkovitz, Iovieno, Ameral, Nelson and Papakostas26, Reference Mallery, MacLeod, Allen, McLean-Veysey, Rodney-Cail and Bezanson47]. To mitigate this risk, it is important to reiterate that the TRLLD designation only applies after two adequate antidepressant treatments have been unsuccessful, and that there are still multiple possible and potentially effective therapeutic avenues available [Reference Pozuelo Moyano, Bautista, Ibarra, Mueller, von Gunten and Vandel14, Reference Houix, Lepetit, Arbus, Armand-Branger, Bonin-Guillaume and Calvet35] when following an evidence-based, guideline-informed sequential treatment approach.

In terms of psychotherapy, 71% of experts indicated that non-response to psychotherapy should not count as a failed treatment in the categorical definition of TRLLD. Although this met the predefined consensus threshold, it was one of the weakest consensus findings and did not remain robust in the sensitivity analyses. However, these analyses should be interpreted cautiously, as most panellists with formal psychotherapy training were retained for the 2nd SR. In the proposed definition, psychotherapy non-response was not retained as a core definitional criterion, but its role should remain open to further evaluation and may be documented within the broader operational staging framework. The lack of consensus also likely reflects the limited and heterogeneous evidence base specific to TRLLD, which complicates operationalisation (e.g., modality, intensity, delivery format). It may also be influenced by variability in access to and delivery of psychotherapeutic services for older adults across health systems. Notably, scientific evidence of LLD emphasises the importance of combining pharmacotherapy with evidence-based psychotherapies wherever possible, while acknowledging that research into psychotherapy for TRLLD remains limited [Reference Pozuelo Moyano, Bautista, Ibarra, Mueller, von Gunten and Vandel14, Reference Ang, Lee, Song, Lee, Cao and Zhang48].

Experts agreed that psychotherapy may be included as one of three failed treatments in an operational definition. Therefore, although treatment resistance is established by the failure of two adequate antidepressant treatments, a history of non-response to psychotherapy should be mentioned explicitly (Table 9).

Most panellists agreed that one should wait approximately 4 weeks to observe the effects of antidepressants. This stands in contrast to literature suggesting that, in patients aged ≥65 years, a longer trial than in general adult populations may be warranted (8 weeks) [Reference Steffens13]. As endorsed by most panel members, waiting 8 weeks is often not feasible in real-world clinical practice, underscoring the need for close monitoring and timely, proactive adjustments when early improvement is absent. In this context, the panel’s preference for week 4 as an optimal early decision point is consistent with the literature on early decision-making in LLD, which suggests that (i) more than 40% of partial responders at week 4 ultimately attain full response by the end of acute treatment; (ii) fewer than 25% of non-responders at week 4 attain full response by week 12; and (iii) early full response is largely sustained through week 12 [Reference Mulsant, Houck, Gildengers, Andreescu, Dew and Pollock49–Reference Kim, Blumberger, Karp, Lenze, Reynolds and Mulsant51]. Accordingly, changing treatment before week 4 may lead to premature discontinuation of potentially effective therapy in too many patients, whereas the absence of at least partial response by week 4 may reasonably prompt treatment modification (e.g., switch or augmentation).

Cognitive impairment and TRLLD

The panel agreed that individuals with dementia should be excluded from the definition of TRLLD and emphasised the importance of determining the aetiology of cognitive decline in older adults with TRLLD. Even if biomarkers are not part of the TRLLD definition, some could be part of the differential diagnosis. Emerging blood-based biomarkers such as plasma p-tau217 or neurofilament light (NfL), markers of specific Alzheimer pathology and non-specific neurodegeneration, respectively, may provide adjunctive evidence to distinguish whether cognitive symptoms in TRLLD reflect a primary neurodegenerative process or are secondary to depression [Reference Food and Administration52–Reference Eratne, Kang, Lewis, Dang, Malpas and Keem55]. However, particularly in the very old, abnormal tau measures may also reflect preclinical pathology and may not map directly onto current symptoms, so results should be interpreted in a clinical context [Reference Ossenkoppele, Pichet Binette, Groot, Smith, Strandberg and Palmqvist56, Reference Rabinovici, Knopman, Arbizu, Benzinger, Donohoe and Hansson57]. Where available and clinically indicated, additional investigations such as amyloid PET imaging for Alzheimer’s disease or cerebrospinal fluid α-synuclein assays for Lewy body disease may further support differential diagnosis and phenotyping [Reference Palmqvist, Rossi, Hall, Quadalti, Mattsson-Carlgren and Dellavalle58]. The lack of consensus on defining patients with TRLLD, no prior depressive history, and comorbid mild cognitive impairment (MCI) as a separate category may reflect the aetiological heterogeneity of MCI in LLD. While cognitive impairment may be at least partly related to the depressive episode in some patients [Reference Morimoto, Kanellopoulos and Alexopoulos59], in others, it may indicate an underlying neurodegenerative process [Reference Yan, Xiang, Wang, Wei, Fan and Du60, Reference Mueller and Pozuelo Moyano61]. Therefore, patients with TRLLD and comorbid cognitive impairment should receive longitudinal cognitive follow-up [Reference Mueller and Pozuelo Moyano61]. From a clinical standpoint, individuals with LLD/TRLLD and cognitive impairment, irrespective of the underlying aetiology, require active treatment, as they may be particularly susceptible to accelerated brain ageing and adverse outcomes [Reference Chan, Chen, Tsai, Bai, Tsai and Cheng10, Reference Diniz, Sibille, Ding, Tseng, Aizenstein and Lotrich62].

Apathy is another frequent and clinically meaningful feature in LLD. In a geriatric cohort, approximately 38% of older adults with depression have clinically significant apathy, and its severity predicts disability independently of overall depressive severity [Reference Yuen, Bhutani, Lucas, Gunning, AbdelMalak and Seirup63]. Importantly, apathy is not a formal diagnostic criterion for a depressive episode in DSM-5 or ICD-10; therefore, its presence or absence should not be used to determine whether the minimum severity threshold (i.e., a moderate depressive episode) is met. Nonetheless, apathy remains clinically relevant in LLD and should be assessed and documented as an associated feature.

Other comorbidities and TRLLD

Some experts questioned whether residual somatic symptoms (e.g., sleep, appetite, fatigue/energy-related symptoms) should count against the TRLLD definition. Ultimately, the statement that “residual symptoms after two or three adequate treatments indicate TRLLD” did not reach consensus, so residual somatic symptoms were not included as a defining criterion. Nonetheless, the clinical management of somatic residual symptoms remains understudied, and further research should define optimal strategies for their assessment and treatment.

Consensus was reached on the need to screen for physical pathology before labelling a case as TRLLD. Importantly, in individuals with high somatic burden, comorbidity is often associated with lower treatment adherence [Reference Rossom, Shortreed, Coleman, Beck, Waitzfelder and Stewart64]. Accordingly, these patients warrant closer adherence monitoring to minimise the risk of misclassifying non-response due to poor adherence as TRLLD. Furthermore, there was consensus to exclude from the definition patients with depressive symptoms occurring in the context of another primary psychiatric disorder (e.g., bipolar disorder or schizoaffective disorder). Further discussions by task force groups are needed to determine whether certain resistance criteria defined for bipolar depression in adults can be adapted to the late-life population [Reference Hidalgo-Mazzei, Berk, Cipriani, Cleare, Florio and Dietch65].

Treatment adherence and tolerance and TRLLD

Our panel agreed that TRLLD assessment should include structured questions about adherence, with input from a proxy such as a partner or carer. This approach can be difficult to implement for socially isolated patients and may also be perceived as intrusive; therefore, it should only be undertaken with the patient’s consent. Nevertheless, it remains clinically relevant, as adherence in older adults with depression is often suboptimal.

Studies in primary care cohorts of older adults show that approximately 13.5% do not commence prescribed antidepressants, 15.2% adhere to treatment insufficiently, and 37.1% do not persist with treatment during roughly the first year [Reference Holvast, Oude Voshaar, Wouters, Hek, Schellevis and Burger66]. In geriatric cohorts, about one-half of older adults achieve an adherence level of at least 80% of days covered [Reference Albrecht, Khokhar, Huang, Wei, Harris and Moyo67]. These findings highlight the clinical importance of structured adherence assessment and, where appropriate, corroborating adherence with proxy input (e.g., a partner or caregiver).

The use of plasma antidepressant levels to monitor adherence was one of the most contentious issues. Some panel members highlighted the limited availability of such testing in most centres and the lack of robust evidence for a direct correspondence between plasma concentrations and adherence, whereas others argued that this relatively inexpensive assay could become more accessible if incorporated into routine practice [Reference Hiemke, Bergemann, Clement, Conca, Deckert and Domschke68].

The British Association for Psychopharmacology’s guidelines for the treatment of depressive disorders emphasise the need for effective strategies to enhance medication adherence [Reference Cleare, Pariante, Young, Anderson, Christmas and Cowen69]. They also note that therapeutic drug monitoring can help assess treatment adherence and lack of efficacy at apparently adequate doses [Reference Cleare, Pariante, Young, Anderson, Christmas and Cowen69].

The authors emphasise that aligning treatment with patient preferences and implementing structured follow-up plans improves both adherence and outcome; that adherence counselling is beneficial, whereas information leaflets alone are not; and that once-daily regimens can be as effective as multiple daily dosing and are associated with better adherence [Reference Cleare, Pariante, Young, Anderson, Christmas and Cowen69]. Psychoeducational interventions can improve outcomes in later life [Reference Higgins, Livingston and Katona70].

Further studies should determine the proportion of patients classified as resistant under the presented definition who exhibit non-adherence or subtherapeutic plasma antidepressant levels.

Social and psychological factors and TRLLD

Finally, the panel reached consensus on the relevance of social determinants in both the definition and the care pathway for TRLLD. Experts agreed that social isolation contributes to poorer treatment outcomes in LLD and that access barriers should be documented and considered when interpreting treatment adequacy and resistance. Costs and insurance coverage limitations restrict pharmacological access in LLD, affecting prescription initiation, continuity, and adherence [Reference Donohue, Zhang, Aiju, Perera, Lave and Hanlon71, Reference Fusco, Sils, Graff, Kistler and Ruiz72]. These factors are contextual determinants relevant to research and clinical implementation, but should not be considered core criteria for defining TRLLD. Addressing these barriers, through facilitated access to medications, financial and logistical support, and collaborative-care models embedded in primary care, is essential. Furthermore, social isolation can restrict access to evidence-based treatments –including pharmacotherapy, psychotherapy, or tele–mental health– and may shape how TRLLD is identified in routine clinical practice. Despite recent improvements in digital connectivity, substantial access gaps persist among socially disadvantaged, rural, and homebound older adults, limiting engagement in remote interventions and exacerbating inequities [Reference Mao, Tam, Xu, Osborn, Sheffrin and Gould73, Reference Choi, Choi and Marti74].

Sensitivity analysis of consensus findings

A best-case/worst-case sensitivity analysis was performed to assess the potential impact of the six experts who did not complete the 2nd SR. This analysis showed that several items close to the 70% consensus threshold were sensitive to assumptions regarding non-responders. These findings do not invalidate the 2nd SR results, which were based on the prespecified Delphi sample of 24 respondents, but they should be considered when interpreting the robustness of individual items. Full item-level results are provided in Supplementary Tables S2 and S3.

Strengths and limitations

This study applies a rigorous Delphi methodology aligned with CREDES guidance (Supplementary Table S1) [Reference Jünger, Payne, Brine, Radbruch and Brearley25], using iterative SR, prespecified consensus thresholds, and structured feedback between SR. The panel was international and multidisciplinary, providing broad clinical input; however, geographical representation was uneven, with a substantial proportion of participants from France, England, and Switzerland. This geographical imbalance may limit the representativeness of the panel and should therefore be considered when interpreting the findings. Item generation was anchored in a comprehensive literature review and refined through open-ended responses (1st SR), enhancing content validity. The process achieved substantial agreement (consensus on 72.4% of closed items) and produced operational outputs, a consensus definition, and a decision algorithm that are readily applicable to practice.

In terms of limitations, the panel included only one psychologist and no psychosocial practitioners, which may have reduced the robustness of consensus on psychotherapy-related items and calls for cautious interpretation in the absence of broader specialist input. This should, however, be considered alongside the fact that the 2nd SR included five psychiatrists from Germany and Switzerland, where the official specialist designation explicitly combines psychiatry and psychotherapy [75, Reference Hohagen and Berger76]. Nevertheless, broader multidisciplinary input, particularly from clinical psychologists and psychotherapy specialists, should be considered in future revisions of the TRLLD definition. More generally, and inherent to the Delphi approach, these findings reflect expert consensus rather than empirical testing; consensus represents agreement among panellists and does not necessarily indicate an objectively “correct” position, while persistent disagreement remains informative. Therefore, future studies should evaluate the proposed TRLLD criteria in independent samples by examining their feasibility in routine practice, inter-rater reliability, and clinical validity. The high proportion of closed questions in the 2nd SR may have constrained expression of nuance, and anchoring effects may have arisen because panellists viewed summaries from previous SR. To mitigate this issue, a free-text field for open comments was included in each category. Another limitation is that not all non-pharmacological approaches were considered in shaping the definition. Some, such as psychotherapy, were considered in the Delphi process. However, other interventions – like physical exercise – with a growing body of evidence supporting their role in TRLLD [Reference Krell-Roesch, Syrjanen, Kremers, Barisch-Fritz, Krafft and Johnson77–Reference Knochel, Alves, Friedrichs, Schneider, Schmidt-Rechau and Wenzler79], were not specifically included.

Finally, while the panel achieved consensus on the importance of adherence monitoring and age-specific pharmacokinetic considerations, practical implementation strategies – including the role and timing of therapeutic drug monitoring, approaches to managing complex drug–drug interactions, and systematic assessment of sickness behaviour in the context of medical comorbidity – were not fully operationalised and require further empirical investigation.