Major Depressive Disorder (MDD) is among the most prevalent and disabling psychiatric conditions worldwide (1). Despite numerous available evidence-based pharmacological treatments, a substantial proportion of patients do not achieve satisfactory outcomes and develop treatment-resistant depression (TRD) (2, 3). Compared to MDD, TRD is associated with higher rates of psychiatric and physical comorbidities, greater functional impairment, increased likelihood of hospitalization, longer hospital stays, higher healthcare costs, and an elevated risk of suicidality (4–11).

Despite its clinical relevance, a consensual operational definition and diagnostic criteria for TRD are lacking (12–14). There is no solid evidence validating any meaningful cut-off based on the number of treatment failures for defining a binary TRD phenotype and a spectrum of resistance seems more plausible (15, 16). Yet, according to the European Medicines Agency (EMA), TRD is considered when treatment with at least two different antidepressant agents (of the same or a different class) of adequate dosage and duration shows lack of clinically meaningful improvement (17). Based on this statement, TRD is frequently defined as failure to respond to two adequate antidepressant (AD) trials. Using this definition, approximately 30% of patients with MDD develop TRD (2, 6).

However, not all responders finally achieve the ultimate therapeutic goal, which is clinical remission, i.e. near-absence of depressive symptoms (18). Therefore, if TRD is defined as lack of clinical remission rather than lack of response after two AD trials, its estimated prevalence rises considerably. For example, in the STAR*D study (19), after the first two trials around 45% of patients cumulatively failed to reach remission (20). Full restoration of psychosocial and cognitive functioning (functional recovery), recently recognized as the optimal goal of treatment, typically takes much longer to achieve, and its measurement varies widely across settings (21). Therefore, clinical remission, signifying the end of the depressive episode, appears as a reasonable preferable clinical anchor on which definition of resistance should be based (22).

Apart from binary definitions of TRD, such as lack of response or remission after two AD trials, which are often used in clinical practice as they are simpler and adopted by regulation authorities, dimensional and staging models have also been proposed to describe multiple dimensions of treatment resistance or gradations in its severity, respectively, rather than simply presence or absence of treatment resistance (13, 14, 23). Staging frameworks include the Thase and Rush model (24), which describes five stages of resistance depending on the pharmacological AD classes sequentially failed and assuming a hierarchy of ADs, and the European Staging Model (ESM), proposed by the Group for the Study of Resistant Depression (GSRD), which describes three main stages of resistance depending on the number and duration of failed AD trials (25, 26).

The Maudsley Staging Method (MSM) is both a dimensional and a staging tool to assess treatment resistance in MDD (27). In addition to AD failures, MSM also rates failure of any augmentation strategies and electroconvulsive therapy (ECT). Unlike earlier staging models, MSM uses lack of remission rather than lack of response to define treatment failures (22). Moreover, it bases assessment of resistance not only on treatment failures but also two additional dimensions, duration of current episode and baseline depression severity. The total MSM score also allows for the classification of patients into mild, moderate or severe stages of resistance. The model has been validated in multiple studies, demonstrating its predictive value for poor prospective clinical outcomes, including increased functional impairment, time spent with depression and lower chance of remission (22, 28–32). By integrating both treatment intensity and clinical burden (depression severity and episode duration), this remission-based dimensional staging model provides additional benefits for treatment resistance quantification both in research and clinical settings.

Despite growing interest in TRD, there is a notable lack of longitudinal, real-world studies following up TRD patients to remission in naturalistic settings. Most existing studies prioritize treatment response as the primary outcome and often have limited follow-up durations, rarely exceeding two years. Regarding MSM, it has been applied only at random snapshots, e.g. at hospital discharge, to predict short- or long-term outcomes. To date, no studies have used MSM to quantify the full continuum of treatment resistance across the Major Depressive Episode (MDE) ending at remission, i.e. episode-wide. Capturing the total burden of treatment failures and time elapsing from episode onset to remission as well as baseline clinical severity in a dimensional framework, such as an episode-wide version of MSM, produces an accurate, continuous index of resistance observed over the whole episode. This index is not an estimate or prediction but a post-hoc record of all treatment efforts required and time needed to reach remission given the initial severity. Therefore, it can serve as a gold standard to compare with yet non-validated categorical classifications, such as remitters, non-remitters or TRD. Furthermore, it might be tested for correlation with candidate biological, clinical or psychological predictors.

This prospective observational study included adult patients with a moderate to severe MDE in the context of MDD who were naturalistically followed up at a specialized tertiary-care center aiming at remission. It was part of the ATLADIS (AThens Longitudinal Affective DIsorders Study) prospective study also including patients with bipolar disorder (33, 34). Using the GSRD/ESM categorical TRD definition but adopting a remission-based framework, patients were categorized as 1^st AD trial remitters, 2^nd AD trial remitters, TRD remitters and TRD non-remitters (categorical outcomes). Episode-wide MSM (EW-MSM) was rated at the end of the MDE (on remission, if achieved) to produce a continuous score of resistance of the MDE from its beginning to its end. The study aimed to:

This study is, to our knowledge, the first to apply EW-MSM, a dimensional score of treatment resistance recorded over the whole depressive episode, from episode onset to remission, as previous research only employed MSM at randomly timed snapshots, such as at hospital discharge, to predict short- or long-term outcomes. This score summarizes most of the clinical and treatment-related information that is required for treatment resistance quantification and that was collected during our naturalistic trial. It was, therefore, considered as a gold standard against which typical remission-based categorical outcomes were compared. Comparison between these two measures of resistance focused on three targets: 1) which were their clinical and treatment-related predictors, 2) how well the two measures predicted depression improvement to remission, and 3) how the two measures were related with each other.

The prevalence of TRD using our remission-based categorical definition was 37.5% (100 TRD patients out of 267 in total). This compares to around 45% of patients cumulatively failing to achieve remission at the second step of the STAR*D study, in which treatment options were more limited, available only after the first step and sequenced in a structured way (20). Using their response-based definition, the GSRD recorded a TRD prevalence of 50.7% and 40.6% in the TRD-I (47) and TRD-III (45, 48, 49) cross-sectional studies, respectively, and 54.5% in the TRD-II (50) prospective study. In summary, discrepances across cohorts may be attributed to different recruitment strategies, settings, treatment procedures followed and TRD definitions employed.

EW-MSM and categorical outcomes were predicted by shared and distinct sets of clinicodemographic variables in multivariate models. Baseline depressive severity and lifetime OCD comorbidity emerged as significant predictors for both EW-MSM and categorical models, while current psychotic features were associated with EW-MSM outcomes only, and family history of MDD in first-degree relatives was significant only in the categorical framework. Baseline depression severity was more strongly associated with EW-MSM than with categorical outcomes, since it is reflected by its severity dimension. Baseline depression severity also emerged as a key predictor of resistance in STAR*D findings, showing that higher initial symptom scores were associated with reduced likelihood of remission at the second step (51), and in GSRD results (25), that similarly identified severity as a robust correlate of resistance in TRD-I (47), TRD-II (52) and TRD-III (53) studies. Of note, mean R-MADRS was 40.6 in our tertiary-care study, with >80% of remitters being hospitalized during their current MDE, while in TRD-II and TRD-III studies mean R-MADRS was 31.5 and 34.1, respectively (48, 50). Lifetime OCD comorbidity (in 9.4% of remitters in our study) was also identified as predictor of resistance at the STAR*D second step (51); in the TRD-III study, the prevalence of comorbid OCD was very low (1.65%) and, although linked with higher suicidal risk and polypharmacy, it was not associated with resistance (54). Current psychotic features (in 19.3% of remitters), a predictor embedded within the MSM severity dimension, were consequently strongly linked with EW-MSM scores in our cohort, in agreement with the TRD-III study (53), where prevalence of psychotic features was 10.9% (48). The association between family history of MDD (in 28.4% of remitters) and categorical outcomes corroborates similar findings in the TRD-II study (52) pointing to the heritable vulnerability of TRD (55). GSRD studies have identified additional clinical predictors of resistance, including number of hospitalizations, suicidality, recurrent episodes, early age of onset, comorbid anxiety disorders and anhedonia/melancholic features (47, 48, 52, 53, 56–58). These factors were not replicated in our study, which may be resulting from our remission-based definition as opposed to response-based frameworks, as well as the single tertiary-care center design and limited power of our sample.

In our study, remitters received a median of 2 ADs per episode, rising to 4 among TRD remitters. Nearly half (49.4%) received AD combinations and 83.3% underwent augmentation during their current MDE; respective percentages in TRD remitters were 84.8% and 92.4%. These rates indicate substantial treatment intensity and are broadly consistent with findings from the GSRD studies. In TRD-I, 18.7% of TRD patients had three and 9.2% four ADs during the index episode (47). In TRD-III, 60.6% received combination or augmentation regimens (45), with 29.5% on AD combinations and augmentation often involving antipsychotics (25.7%) or mood stabilizers (11.3%), while mean number of psychiatric drugs administered reached 2.18. Our cohort showed high pharmacological exposure, with mean cumulative fluoxetine equivalent doses of 89.9 mg across all trials versus 39.9 mg in TRD-III (45). This difference likely reflects the longer follow-up of our study subjects followed up to remission, demanding more AD trials and more intensive dose optimization. Treatment-related characteristics were unsurprisingly strongly associated with both EW-MSM and categorical outcomes. Longer episode duration predicted higher resistance in both EW-MSM and categorical models. Moreover, in multivariate models EW-MSM was associated with the total number of ADs received while categorical outcomes with the use of AD combinations. Additionally, EW-MSM was significantly predicted by the use of augmentation strategies, since these are included as a separate indicator of resistance. Overall, both clinicodemographic and treatment-related characteristics predicted EW-MSM more efficiently than categorical outcomes, underscoring its superior ability to reflect the multidimensional aspects of resistance (clinical burden, treatment intensity).

The hierarchical regression analysis demonstrated the superiority of EW-MSM over categorical outcomes in capturing symptom improvement from baseline to remission. This was mainly attributed, in a sensitivity analysis, to the fact that EW-MSM, through its severity dimension, was more strongly associated with baseline depression severity than categorical outcomes. Furthermore, EW-MSM can better reflect several other key aspects of resistance while categorical outcomes only rely on discrete AD trial counts. As a result, EW-MSM provides a more accurate and comprehensive index of resistance, which also captures therapeutic gain more efficiently than categorical outcomes.

Comparative analyses of EW-MSM dimensions across categorical outcomes showed a clear gradient of resistance, increasing from 1^st to 2^nd AD trial and TRD remitters (C1>B>A), only for the Antidepressants dimension, as number of ADs in EW-MSM is highly correlated with number of (monotherapy or combination) AD trials in categorical outcomes. Other EW-MSM dimensions were not equally discriminative. EW-MSM Duration clearly separated TRD remitters from early remitters but not among the two early remitters subgroups. EW-MSM Severity performed poorly and only when examined categorically while EW-MSM Augmentation clearly separated only among the most distant categories. These findings suggest that typical categorical outcomes are clearly separated among them only by the number of ADs received, based on which they are defined, but less clearly by several other key aspects of resistance, such as clinical severity, episode duration, and augmentation strategies. Yet, EW-MSM total score was strongly though not perfectly correlated with categorical outcomes and performed as well as the Antidepressants dimension, clearly distinguishing between all three categories on the same gradient, maybe because this dimension contributes up to 5 points to the total score.

The ROC curve analyses demonstrated that EW-MSM discriminated effectively (AUC≥0.80) only between the most distant categories (1^st AD trial remitters vs. TRD remitters; AUC = 0.91) but not among adjacent categories. The suboptimal discriminative efficiency of EW-MSM between adjacent categories, especially among the first two, arises from the fact that our sample included patients with high severity intensely treated early on. Even in 1^st AD trial remitters, 20% received AD combination and 75% received augmentation. Most remitters (76.8%) had moderate resistance by MSM cut-offs, making it more difficult to distinguish subgroups in them. As noted above, number of ADs tightly aligned with outcome categories, but other EW-MSM dimensions showed higher dispersion across them, resulting in poor discrimination between adjacent categories in ROC analyses. As has been proposed, MSM could become more granular and increase its discriminative efficiency by incorporating the number of augmentation strategies and AD combinations used (13).

Comparisons in ROC analyses were also extended to include combined groups. As pharmacological trials in TRD patients are sparse (15, 59), treatment guidance often relies on trials in patients failing to remit after at least one adequate AD trial (60). Psychiatric pharmacogenetic studies also often focus on non-remitters after one AD trial (61). Therefore, comparing 1^st AD trial remitters with all other groups is of clinical relevance; discrimination among them was fair in our study (AUC = 0.815). When 1^st and 2^nd AD trial remitters were combined and contrasted with TRD remitters, the discriminative efficiency was even higher (AUC = 0.865), suggesting that TRD remitters are quite distinct from early remitters and lending validity to the TRD definition. Discriminative cut-offs identified in ROC analyses included EW-MSM scores of 8 and 9 or higher. The former is useful in distinguishing 1^st AD trial remitters from all other groups combined or TRD remitters; the latter in distinguishing between TRD and early remitters. Both cut-offs lie in the moderate resistance range of scores (27).

Overall, findings from the comparisons between EW-MSM and categorical outcomes on our three aims are summarized as follows: 1) EW-MSM score was significantly associated with both clinicodemographic and treatment-related predictors, overlapping with, but also extending beyond those associated with categorical outcomes. Both approaches converged on baseline severity, current MDE duration, lifetime OCD comorbidity and number of ADs or AD combination as significant predictors in multivariate models, while EW-MSM was additionally predicted by current psychotic features and use of augmentation strategies. However, predictions were always more efficient for EW-MSM than for categorical outcomes, demonstrating its superior ability to reflect the clinical burden and treatment intensity aspects of resistance. 2) EW-MSM was more strongly associated with baseline severity than categorical outcomes and consequently predicted depression improvement to remission more efficiently. 3) EW-MSM was strongly though not perfectly associated with categorical outcomes. Only the Antidepressants dimension and the total score showed a clear gradient of resistance across all three outcome categories. As a result, EW-MSM effectively discriminated between the most distant categories (1^st AD trial remitters vs. TRD remitters) but not among adjacent categories. This does not necessarily point to an EW-MSM drawback; categorical outcomes defined by number of AD trial failures were not as distinct regarding other key aspects of resistance captured by EW-MSM and were therefore not adequately distinguishable by it.

This study has notable strengths. To our knowledge, this is the first application of MSM across the entire episode, whereas previous research had employed it at random snapshots, such as at hospital discharge (28–32). We derived a continuous score of episode-wide resistance, capturing both clinical burden (symptom severity, episode duration) and treatment intensity (AD trials, AD combination and augmentation). This was used as a reference against which remission-based categorical classifications were compared. The longitudinal, naturalistic design of the study mirrors real-world clinical challenges related to following up TRD patients to remission. We prioritized clinical remission over symptom response, aligning with contemporary treatment objectives emphasizing remission and functional recovery beyond remission, but recognizing that among them clinical remission is the most feasible and finally the preferred endpoint to define a successful clinical outcome (22).

Several limitations should also be acknowledged. First, the cohort did not include patients receiving ECT or newer interventions indicated in TRD, such as esketamine, which may limit generalizability to all therapeutic strategies. Moreover, unavailability of ECT did not allow scoring of the full EW-MSM tool in this cohort. Total scores might be higher, particularly for TRD patients, if ECT was available. Second, MSM does not capture non-pharmacological interventions, like psychotherapy or rTMS; consequently, patients receiving psychotherapy were excluded from our study to maintain consistency in applying MSM, though at the cost of external validity. Third, no objective measures of adherence, such as AD plasma levels, were used; suboptimal adherence may lead to pseudo-resistance (12). Fourth, the timing of episode onset and baseline depression severity were retrospectively determined and, therefore, potentially subject to recall bias. Finally, this study was conducted at a single tertiary-care center and included intensely treated patients with high severity, introducing selection bias; therefore, replication in other clinical settings is needed to ensure the findings are representative of broader clinical populations.

Our findings suggest several directions for future clinical and translational research. Following up MDD (incl. TRD) patients to remission in clinical research settings and using EW-MSM to measure treatment resistance could help identify data-driven cut-offs of resistance through latent class/profile or cluster analyses of EW-MSM dimensions, possibly combined with patient-reported outcomes. These cut-offs could prove of greater clinical utility than typical categorical classifications and be used for drug discovery or stratification in clinical trials. Furthermore, EW-MSM total score or specific dimensions might be tested, in an unbiased and hypothesis-free way (i.e. without relying on arbitrary cut-offs based on AD trial failures), for correlation with putative biological (e.g. inflammatory or oxidative stress biomarkers, neuroimaging data, polygenic scores, etc.), clinical (e.g. psychiatric or medical comorbidity) or psychological (e.g. personality profile, cognition, childhood trauma, coping strategies, etc.) predictors. Timing should be considered in these analyses, facilitating the identification of stage-specific biomarkers for use in precision psychiatry frameworks. In such pragmatic studies follow-up can be long to accurately reflect routine care but could be justified by the expected benefits. Future studies should evaluate EW-MSM scoring in larger and more diverse populations to confirm applicability and strengthen generalizability. Finally, feasibility could be enhanced by applying the EW-MSM approach in electronic health records databases or patient registries.

In conclusion, our study highlighted the clinical importance of prioritizing remission over response as the therapeutic target and for defining resistance, showing that remission is feasible even among patients with TRD, yet after more treatment attempts. EW-MSM produced a continuous score of episode-wide resistance, capturing both clinical burden (symptom severity, episode duration) and treatment intensity, which was used as a reference against which remission-based categorical outcomes were compared. EW-MSM and categorical outcomes were predicted by shared but also distinct sets of clinicodemographic and treatment-related variables. However, EW-MSM reflected clinical burden, incl. baseline depression severity, and treatment intensity aspects of resistance more strongly than categorical outcomes and consequently predicted depression improvement to remission more efficiently. EW-MSM scores followed a clear gradient of resistance across categorical outcomes and effectively discriminated between the most distant categories but not among adjacent ones. Overall, our results demonstrate the added value of using EW-MSM and support its use in clinical and translational research to identify the clinical, biological and psychological correlates of resistance.