Association between initial treatment strategy and progression to difficult-to-treat rheumatoid arthritis (D2T-RA): evidence from a 20-year-old real-world cohort

Abstract

Background/objectives

Despite major advances in rheumatoid arthritis (RA) management, a substantial proportion of patients evolve to difficult-to-treat RA (D2T-RA). We estimated the probability of progression to D2T-RA and assessed whether the initial choice of biological/targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs) is associated with this risk.

Methods

We retrospectively analysed 675 patients with established RA from the first consultation through all b/tsDMARD lines to the last follow-up or D2T-RA classification. Kaplan-Meier, log-rank and multivariable Cox regression analyses were performed to identify factors associated with D2T-RA progression.

Results

Of the 675 patients (mean follow-up of 217.1±131.16 months), 126 (18.7%) fulfilled the D2T-RA criteria. Tumour necrosis factor inhibitor (TNFi) accounted for most first-line and second-line treatments (81.0% and 53.2%). In the post-2000 subcohort (498 patients), no significant difference in D2T-RA risk was observed between TNFi and non-TNFi starters nor between TNFi-cycling or other mechanism of action (MoA)-switching strategies. Glucocorticoid exposure was independently associated with increased risk (HR 2.47, 95% CI 1.38 to 4.42) and delayed b/tsDMARD initiation with decreased hazard (HR 0.990, 95% CI 0.98 to 0.996).

Conclusion

In this observational cohort, initial MoA choice was not associated with D2T-RA progression. Disease severity markers appeared more prominent than treatment strategies in driving refractory evolution. Validation in a prospective/randomised setting is required before potential transition into clinical practice.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• The difficult-to-treat rheumatoid arthritis (D2T-RA) classification criteria have been recently defined by European Alliance of Associations for Rheumatology (EULAR). The characteristics of patients with D2T-RA seem to be driven by high disease activity. It is currently unclear whether the progression to D2T-RA is intrinsically determined or whether some therapeutic strategies could minimise the risk of developing D2T-RA.

WHAT THIS STUDY ADDS

• Progression to D2T-RA status was not significantly affected by whether a tumour necrosis factor inhibitor (TNFi) or non-TNFi was used as the first targeted biological/targeted synthetic disease-modifying antirheumatic drug in our study cohort.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our findings give further support to current guidelines that leave physicians the choice across all types of mechanism of action as first targeted therapies (based on patient profile, price, availability, etc). Confirmation in a prospective/randomised setting will be informative.

Introduction

Rheumatoid arthritis (RA) is a chronic inflammatory disease characterised by persistent synovitis, progressive joint damage and systemic manifestations. In patients with unsatisfying response (or intolerance to conventional synthetic disease-modifying antirheumatic drugs (csDMARDs)), targeted therapies (biological/targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs)) with different mechanisms of action (MoA) become part of the standard clinical practice. These include tumour necrosis factor inhibitors (TNFi), interleukin-6 receptor inhibitors (anti-IL6R), T-cell costimulation modulators (anti-CTLA4), B-cell-depleting therapies (anti-CD20) and Janus kinase inhibitors.

However, a significant proportion of patients with RA do not achieve control of the disease and need multiple lines of treatment.^1–5

Patients with insufficient response to different targeted therapy led to the conceptualisation of difficult-to-treat RA (D2T-RA).⁶ The D2T-RA definition includes a moderate to high disease activity despite at least two b/tsDMARDs with two different modes of action. The definition also includes other non-inflammatory parameters such as chronic pain, fatigue, comorbidities, related toxicity and often prolonged exposure to glucocorticoids (GCC).^7–9 It can apply to up to 20% of patients with RA, according to various cohorts, and is associated with a significant medical and socioeconomic burden.¹⁰

Previous studies identified several baseline characteristics associated with later progression to D2T-RA, such as a younger age, higher disease activity, concomitant fibromyalgia, higher disease burden, lower quality of life and higher doses of GCC.^{11 12}

Yet, it is currently unclear whether the progression to D2T-RA is intrinsically determined, or whether some ‘actionable’ therapeutic strategies could minimise the risk of developing D2T-RA. A key candidate factor is the MoA of the first targeted therapy (b/tsDMARDs) used. Although the most recent American College of Rheumatology/European Alliance of Associations for Rheumatology (ACR/EULAR) guidelines allow the use of any b/tsDMARDs after failure of first-line therapy,^{13 14} TNFi has historically been the most frequent first choice. In the present work, we aimed to investigate whether the choice of first-line targeted therapy (TNFi vs non-TNFi) could impact the risk of progressing to D2T-RA.

Methods

Design of the study

We performed a retrospective, single-centre cohort study (UCLouvain Brussels RA cohort) including patients with RA fulfilling 2010 ACR/EULAR or 1987 ACR criteria who had received at least one b/tsDMARD. Patients were followed longitudinally until 1 August 2024 or until their last available clinical visit, whichever occurred first. Time to D2T-RA was defined as the interval from RA diagnosis to the visit at which patients met the EULAR definition of D2T-RA.

Definition of study cohorts and timing of b/tsDMARD exposure

The identification of patients with D2T-RA was based on the simultaneous presence of the following criteria:

1. Treatment in accordance with EULAR recommendations, with failure of at least two b/tsDMARDs (with different MoAs) after failure of csDMARDs (unless contraindicated).

2. Signs suggestive of active/progressive disease, defined as ≥1 of the following:

   • Moderate to high disease activity (Disease Activity Score 28 joints (DAS28)–C-reactive protein >3.2 or Clinical Disease Activity Index (CDAI) >10).

   • Inability to taper GCC below 7.5 mg/day (prednisone equivalent) over a prolonged period, in accordance with EULAR criteria.

   • Rapid radiographic progression: even though it was not systematically assessed in our cohort, it was not reported as a reason for treatment change.

Patients who had received treatment for less than 3 months with a second class of b/tsDMARDs were not considered to have truly tried two classes, as this duration is insufficient to assess efficacy. These patients were therefore excluded from the initial analysis.

Data collection

As part of routine clinical practice, disease activity was recorded using the Tender and Swollen Joint Count 28, the DAS28, the Simplified Disease Activity Index and CDAI. In addition, patient-reported outcomes were obtained from each patient, including visual analogue scales (VAS) for patient global assessment (VASp), fatigue (VASf) and pain (VASpain), as well as the physician’s global assessment (VASm; 0–100 scale) and the Health Assessment Questionnaire (HAQ; 0–3 scale). At each time point, treatment was recorded, including GCC dose (prednisone equivalent, mg/day), csDMARDs, bDMARDs and tsDMARDs, together with the reason for discontinuation. Hands and feet plain X-rays were performed at baseline for each patient to detect radiographic damage as a binary variable (yes/no), including bone erosions (BE) and/or joint space narrowing.

Outcomes of the study

The primary objective of this study was to determine the proportion of patients with RA developing the EULAR definition of D2T-RA within our hospital-based cohort, and to identify whether the first line of treatment might influence the evolution towards that phenotype, distinguishing those who initially received TNFi from those treated with another MoA.

Secondary objectives were:

• To evaluate the time to D2T-RA using survival analysis and to assess the potential impact of the therapeutic era and baseline characteristics on the risk of progression to a D2T phenotype.

• To analyse treatment sequences among patients with D2T-RA, including reasons for treatment changes.

Statistical analysis

Descriptive statistics of clinical and demographic characteristics, as well as disease activity and treatments, were reported as frequency (for qualitative variables) or as the mean±SD or median (IQR) when appropriate (for quantitative variables). To assess normality, the Shapiro-Wilk test and the Kolmogorov-Smirnov test were performed. An independent t-test was used for the comparison of normally distributed continuous variables. Variables with non-normal distribution were compared with the Mann-Whitney test or the Kruskal-Wallis test. Fisher’s exact test or χ² test was used for categorical variables. Time-to-event data were analysed using survival methods. Survival curves were estimated by the Kaplan-Meier method, and differences between groups were assessed with the Mantel-Cox (log-rank) test.

To evaluate the association between clinical and treatment variables and the time to development of D2T disease, univariate and multivariate Cox proportional hazards regression models were applied to identify independent predictors. Classical factors associated with RA severity and poor outcomes, including sex, age, serological status (rheumatoid factor (RF) and anti-citrullinated protein antibody (ACPA) positivity), smoking status and baseline erosive disease, were included a priori based on their established clinical relevance. In addition, variables showing an association with the outcome in univariate Cox analyses were considered for inclusion in the multivariate model. A p value <0.05 was considered significant.

To minimise bias related to differential access to biologic therapies over time, all treatment pathway analyses were restricted to patients diagnosed after 1 January 2000. This approach ensured that all patients were managed during an era in which b/tsDMARDs were available.

Missing data were handled using a complete-case approach. Patients with missing values in any variable included in the multivariable Cox model were excluded from that analysis.

Results are presented as HRs with 95% CIs.

The proportional hazards assumption was verified by inspection of Schoenfeld residuals. All analyses were performed using R software (V.4.4.1; R Foundation for Statistical Computing) employing the following packages: survival, broom, tidyverse, ggplot and survminer.

Results

We included 675 patients with RA, with a mean time of follow-up of 217.1±131.16 months. 126 patients (18.7%) fulfilled the criteria for D2T-RA. Demographic and disease features of the whole population, D2T-RA subgroup and non-D2T-RA subgroup are summarised in table 1. Patients with D2T-RA were younger, with a longer disease duration and higher disease burden (HAQ). As expected, patients with D2T-RA also more commonly received GCC therapy before b/tsDMARDs. The values in bold indicate statistically significant p-values.

Table 1. Demographic and disease features.

	Whole cohort	D2T-RA	Non-D2T-RA	P value
Number of patients	675	126	549	–
Age at diagnosis	41.8±13.5	39.5±12.8	42.3±14.9	0.032
Sex (female, n (%))	(517) 76.6%	77.0%	76.5%	0.908
Baseline bone erosions (n, %)	(376) 57.3%	63.9%	55.8%	0.196
Smoker (n, %)	(152) 26.0%	26.8%	25.8%	0.839
Time of follow-up (months)	217.1±9.8	243.1±10.7	211.2±5.6	0.013
Time to D2T-RA (months)	NA	158.1±10.3	NA	NA
DAS28-CRP at first b/tsDMARDs	4.74±0.05	4.97±0.16	4.69±0.05	0.069
HAQ at first b/tsDMARDs	1.34±0.04	1.60±0.12	1.30±0.04	0.004
Time to first csDMARDs (months)	23.4±2.4	23.2±5.4	23.5±2.7	0.959
Time to first b/tsDMARDs (months)	135.4±57.3	70.5±7.8	150.1±70.2	0.590
GCC before b/tsDMARDs (n, %)	67.2%	82.8%	63.8%	<0.001
MTX before b/tsDMARDs (n, %)	67.2%	85.4%	88.4%	0.349

Bold indicates significant p values.

b/tsDMARDs, biological/targeted synthetic disease-modifying antirheumatic drugs; csDMARDs, conventional synthetic disease-modifying antirheumatic drugs; DAS28-CRP, Disease Activity Score 28 joints–C-reactive protein; D2T-RA, difficult-to-treat rheumatoid arthritis; GCC, glucocorticoids; HAQ, Health Assessment Questionnaire; MTX, methotrexate.

Association between b/tsDMARD selection and risk of progression towards D2T-RA

The primary objective of this study was to determine whether the MoA of the first line and second line of b/tsDMARDs was associated with evolution towards D2T-RA.

As expected, TNFi accounted for most first-line and second-line treatments (81% and 53.2%, respectively) (figure 1A). We found no difference in the probability of evolving towards D2T-RA between patients who started a TNFi as first targeted therapy (n=366) compared with those who started a non-TNFi targeted therapy (n=132) (figure 1B). A similar conclusion emerged when analysing the first two treatment lines together (figure 1C): the initial therapeutic strategy (TNFi → TNFi vs TNFi → other MoA) did not influence the risk of progression to D2T-RA. Although the TNFi → TNFi group had a longer follow-up (181 (116) months vs 164 (108) months, p=0.17) and received more b/tsDMARDs overall (3 (2) vs 2 (2), p<0.0001), the proportion of patients who became D2T-RA was comparable (31.4% vs 26.6%, p=0.83), and time to D2T-RA did not differ significantly (136 (124) months vs 105 (102) months, p=0.75) between these two subgroups.

Figure 1. (A) Heat map reporting each MoA for every line of treatment. (B) Kaplan-Meier curve comparison of D2T-RA evolution between first-line treatment strategy (TNFi vs other bDMARDs). (C) Kaplan-Meier curve comparison of D2T-RA evolution between second-line treatment strategy among patients receiving TNFi as first-line treatment (TNFi ->TNFi vs TNFi -> bDMARDs). (D) Kaplan-Meier curve (no statistical test performed) of D2T-RA evolution between first-line treatment strategy (TNFi vs other bDMARDs). Abbreviations: bDMARDs = biological disease-modifying antirheumatic drugs; D2T-RA = difficult-to-treat rheumatoid arthritis; IL6R = interleukin-6 receptor inhibitor; JAKi = Janus kinase inhibitor; MoA = mechanism of action; TNFi = tumour necrosis factor inhibitor.

Among patients who ultimately developed D2T-RA, those who started with TNFi showed a slower progression to D2T-RA status (figure 1D). Notably, treatment duration of first-line and second-line b/tsDMARDs was similar in patients with and without D2T-RA: 54.1 months (IQR 128.7) vs 57.2 months (IQR 140.4, p=0.92) for first-line therapies and 26.7 months (IQR 106.5) vs 41.0 months (IQR 97.6, p=0.48) for second-line therapies.

b/tsDMARD prescription and reasons for discontinuation

Patients with D2T-RA had received multiple b/tsDMARDs, with 75.4% exposed to three, 50.8% to four and 33.3% to five lines of therapy, and progressively fewer patients receiving higher numbers of treatments (14.2% (6), 6.3% (7), 2.4% (8) and 0.8% (9)). Reasons for discontinuation of first and second b/tsDMARDs were compared between D2T-RA and non-D2T-RA. As reported in figure 2 and online supplemental table 3A,B, patients with D2T-RA mainly withdrew drugs due to primary or secondary failure, while other reasons such as pregnancy and infections were similar between groups. More precisely, loss of efficacy (LoE) and lack of efficacy (LackOE) were the major reasons for first-line b/tsDMARD discontinuation for D2T-RA, especially regarding the comparison with non-D2T-RA (35.1% vs 23.4%, p=0.02, and 30.6% vs 15.3%, p=0.001, respectively). Second-line observation gave a similar result for LoE (31.4% vs 12.2%, p=0.01), while LackOE was comparable (40.5% vs 48.9%, p=0.33).

Figure 2. Reasons for discontinuation of first (A) and second (B) b/tsDMARD lines between D2T-RA (squared columns) and non-D2T-RA (full colour). Reasons for discontinuation up to seventh line of treatment for D2T-RA (C). For 13 patients in the D2T group, reasons for discontinuation were not retrieved. b/tsDMARD, biological/targeted synthetic disease-modifying antirheumatic drug; D2T-RA, difficult-to-treat rheumatoid arthritis. *p=0.02, ***p=0.001, **** p <0.0001.

We confirm that LoE and LackOE were the main reasons to withdraw b/tsDMARDs, and they increased according to the number of b/tsDMARD failures over time.

D2T-RA prevalence and predictors

Over a follow-up of between 3 and 24 years, the mean time to fulfilling D2T-RA criteria was 111.78 months (±72.8 months). The Kaplan-Meier analysis (figure 3) shows the progression to D2T-RA over time; at 50, 100, 150, 200 and 250 months, D2T-RA prevalence was 4.6%, 10.4%, 12.9%, 18.5% and 27.1%, respectively.

Figure 3. Kaplan-Meier curve and forest plot summarising HR for the development of D2T-RA. ACPA, anti-citrullinated protein antibodies; bDMARDs, biological disease-modifying antirheumatic drugs; D2T-RA, difficult-to-treat rheumatoid arthritis; GCC, glucocorticoids; MTX, methotrexate.

The multivariable Cox proportional hazards regression model, restricted to patients diagnosed after 1 January 2000, identified GCC exposure before b/tsDMARDs as an independent risk factor for D2T-RA progression (HR 2.47, 95% CI 1.38 to 4.42, p=0.002). Methotrexate (MTX) exposure before b/tsDMARDs showed a trend towards a protective effect that did not reach statistical significance in this subcohort (HR 0.58, 95% CI 0.33 to 1.04, p=0.06). Additionally, delayed initiation of b/tsDMARDs was independently associated with a reduced hazard of D2T-RA (HR 0.990, 95% CI 0.98 to 0.996, p=0.002). Full univariate results are reported in online supplemental table 1A.

Time-stratified analysis

Among post-2000 patients, those diagnosed after 2010 (n=212, 43.9%) showed a significantly higher probability of progressing to D2T-RA compared with those diagnosed before 2010 (n=286, 56.1%), as illustrated by the Kaplan-Meier analysis (p=0.024, figure 4). In the multivariable Cox model restricted to this subgroup, two variables were independently associated with D2T-RA progression: GCC exposure before b/tsDMARDs (HR 1.72, 95% CI 1.39 to 5.33) and time to first b/tsDMARDs, which remained a strong independent predictor (HR 0.97, 95% CI 0.96 to 0.98)—consistent with findings from the primary post-2000 analysis. Sex, MTX exposure, RF, baseline erosions, age and ACPA were not significantly associated with D2T-RA progression in this subgroup. Full univariate results are provided in online supplemental table 1B. Demographics of pre-2000 and after 2000 cohort are reported in Supplementary Table 2A and 2B.

Figure 4. Kaplan-Meier curve comparison (log-rank test) between patients diagnosed before (purple) and after (orange) 1 January 2010. Forest plot summarising HRs. ACPA, anti-citrullinated protein antibodies; bDMARDs, biological disease-modifying antirheumatic drugs; b/tsDMARDs, biological/targeted synthetic disease-modifying antirheumatic drugs; GCC, glucocorticoids; MTX, methotrexate; RF, rheumatoid factor.

Discussion

In our large retrospective monocentric cohort study, we observed a prevalence of 18.7% of D2T-RA over a mean time of follow-up of 18 years. This proportion is in line with previous observations, reporting prevalence up to 20%.^{15 16}

Patients with D2T-RA were younger, with a more severe functional impairment, according to HAQ scores, and more frequently were exposed to GCC before the first b/tsDMARDs, in agreement with previous findings.^{11 17 18} These results indicate a clear association between initial disease severity, persistent inflammation, GCC dependence and the eventual development of refractory disease. In contrast, other potential predictors—such as baseline erosions and ACPA/RF seropositivity—did not differ significantly between groups, suggesting that their long-term impact may be more limited as opposed to what we have observed in our ERA(Early Rheumatoid Arthritis) cohort.¹² Moreover, early biologic transition and seropositivity (RF and ACPA) were associated with improved short-term to mid-term disease control, suggesting the existence of a therapeutic window during which immunomodulatory strategies may yield maximal benefit. Nevertheless, the existing literature on these factors remains heterogeneous.¹⁸

This study is among the first to specifically evaluate whether the choice of first targeted therapy influences the subsequent development of D2T-RA. To address potential confounding related to historical treatment availability and the structural biases inherent to pre-2000 diagnoses—including non-random treatment allocation and limited access to non-TNFi agents—this analysis was restricted to patients diagnosed after 1 January 2000. In this contemporary subcohort, patients first treated with a TNFi showed a similar long-term risk of fulfilling D2T-RA criteria compared with those initiating a non-TNFi targeted therapy. Consistent results were observed when considering the first two lines of targeted therapy: neither a TNFi-cycling strategy nor an early switch to another MoA significantly affected the likelihood of developing D2T-RA. These findings support current treatment recommendations, which do not prioritise one MoA over another as initial targeted therapy, and instead advocate individualised treatment decisions based on patient characteristics, comorbidities, drug availability and cost considerations.

The apparently slower progression to D2T-RA among TNFi starters within the D2T subgroup should not be interpreted as a protective biological effect. Rather, it likely reflects a definitional artefact: since D2T-RA requires failure of at least two distinct MoAs, patients initiating a non-TNFi may reach this threshold through fewer treatment steps. Consequently, time-to-D2T-RA comparisons between MoA groups are structurally confounded by the outcome definition and are not used to support the primary conclusion; the more robust endpoint remains the cumulative probability of fulfilling D2T-RA criteria, which was comparable between groups and unaffected by this circularity. This underscores a fundamental conceptual distinction between optimising short-term response after biologic failure and preventing long-term refractory evolution, which appears more strongly driven by intrinsic disease biology than by pharmacological sequencing.

Contrary to these findings, recent meta-analyses and systematic reviews have demonstrated that in patients with RA with inadequate response to an initial biologic, switching to a therapy with a different MoA can provide superior clinical outcomes compared with intraclass cycling, particularly in primary TNFi non-responders.¹⁹ These findings were applied in the R4RA study to evaluate whether synovial tissue biomarkers could improve response to rituximab or anti-IL6R after TNFi failure,²⁰ highlighting the mechanistic diversification of RA and reinforcing the need to optimise class switching in severe disease. More recently, a study reported a potential superiority of anti-IL6R over other MoA in this setting,²¹ although this evidence remains preliminary and warrants confirmation in larger or more diverse populations before firm conclusions on optimal MoA sequencing can be drawn.

Our long-term cohort data provide a complementary and more nuanced perspective: while class switching may optimise short-term therapeutic response, early treatment strategy alone may not be sufficient to prevent the development of a refractory phenotype over the long term.

These results question whether MoA influences the disease evolution, suggesting that treatment strategies and physiopathology play a more central role. For example, a faster therapeutic transition may be itself a risk factor for an earlier evolution to D2T-RA, while in other cases, comorbidities and non-inflammatory chronic pain could lead the therapeutic transition.

In our cohort, the analysis of reasons for discontinuation confirms that therapeutic transition in D2T-RA is largely dependent on the efficacy of the treatment. Indeed, primary inefficacy and secondary inefficacy were consistently the principal reasons for drug failure, increasing from 70% at the first line to almost 100% in the sixth line of treatment. More interestingly, LackOE and LoE discriminated D2T-RA and non-D2T-RA, being significantly less frequent in the latter group, resembling the phenotype of the difficult-to-manage RA instead of D2T-RA. This evolution highlights the persistence of refractory inflammation in D2T-RA and the ability of the EULAR definition to identify truly active patients.

Although the number of patients is small and precludes formal analysis, the observed rate of early b/tsDMARD withdrawal is noted as an exploratory observation warranting dedicated investigation. A recent study showed that first-line TNFi induction in early RA was associated with a markedly lower frequency of D2T disease after 5 years,²² and prospective studies are needed to confirm whether induction strategies influence long-term refractory evolution.

In the post-2000 subcohort, the multivariable Cox model, limited by the absence of the inclusion of disease activity, identified GCC exposure before b/tsDMARDs as an independent risk factor for D2T-RA progression, likely reflecting underlying disease severity rather than a direct pharmacological effect. MTX exposure before b/tsDMARDs showed a trend towards a protective effect that did not reach statistical significance. Delayed initiation of b/tsDMARDs was independently associated with a reduced hazard of D2T-RA. This finding should be interpreted cautiously for two reasons. First, given the time-dependent nature of the D2T-RA definition, a longer time to first b/tsDMARDs mechanically compresses the window in which the D2T-RA criteria can be fulfilled. Second, earlier b/tsDMARD initiation may itself reflect greater disease severity, acting as a marker of aggressive disease trajectory rather than a causal determinant of refractory evolution. Overall, initial disease severity and treatment context appear to be stronger determinants of refractory evolution than MoA selection per se. These associations are consistent with previous reports, and their replication in a cohort with up to 20 years of follow-up adds supporting evidence to the existing literature.²³

The development of D2T-RA is to some extent a time-dependent outcome, and the operational definition becomes less informative when projected over very long-time spans.²⁴ The most meaningful analysis should therefore consider the full disease trajectory—from early RA to D2T—within a relatively contained timeframe, as this approach allows for the identification of clearer risk factors, such as disease activity and early BE.¹²

When examining only the past 15 years—a period characterised by the availability of multiple MoAs—a faster progression to D2T-RA was observed among patients with RA onset after 2010 (figure 4). This difference is not attributable to cohort heterogeneity, as both subgroups are otherwise comparable. A plausible explanation lies in evolving treatment strategies: switching to a different MoA—rather than cycling through multiple TNFis, as was more common before 2010—may have increased the likelihood of failing two distinct MoAs, thereby more rapidly satisfying the D2T-RA definitional threshold. This interpretation is consistent with the broader observation that the increasing availability of therapeutic options, while beneficial for individual patients, may paradoxically accelerate the formal classification of D2T-RA by expanding the range of MoA that can be sequentially failed. Time to first b/tsDMARDs remained an independent predictor in this subgroup, further supporting the time-sensitive nature of D2T-RA classification.

Recently, concerns have emerged regarding several pitfalls in the current D2T-RA definition.²⁵ For example, its stability over time is not assured: a patient may enter and exit the D2T category multiple times throughout the disease course. It also appears essential to revisit the current EULAR definition to better distinguish cases of true biological non-response from those driven by behavioural or psychosocial factors, or from an initial misdiagnosis.

This study has several limitations. First, its retrospective design carries inherent risks of selection and information bias. Some components of the EULAR definition—particularly the subjective perception of management difficulty—could not be assessed due to limitations in available documentation. The analysis was conducted in a single university centre, which may limit the generalisability of findings to broader and more diverse populations. Missing data—including GCC and csDMARD dosages—may represent key determinants not captured in our analysis. Although restricting the treatment pathway analysis to post-2000 patients mitigates historical confounding, residual confounding by indication cannot be excluded. Moreover, as already mentioned before, disease activity was not included as a covariate in the multivariate model. Furthermore, as this was designed as an association analysis, results should not be interpreted as evidence of causal treatment effects or equivalence between treatment classes.

Nevertheless, the systematic and comprehensive evaluations conducted, together with the monocentric nature of the cohort, ensured a high level of data quality throughout the long-term follow-up.

Further research is needed to validate early predictive factors of D2T-RA in long-term prospective cohorts and to better understand the biological mechanisms underlying therapeutic resistance. The integration of biomarkers, imaging data and artificial intelligence tools—particularly machine learning approaches—may enhance our ability to anticipate (and potentially prevent) progression towards a D2T phenotype.

It also appears essential to refine the current EULAR definition to better distinguish cases of true biological non-response from those driven by behavioural or psychosocial factors, or from an initial misdiagnosis. From this perspective, we are currently testing an alternative D2T definition (ie, failure of at least three b/tsDMARDs) to more precisely identify real refractory patients and predictors.

Data availability statement

Data are available upon reasonable request.