Abstract
Objectives
The aim of this study was to identify robust predictive markers which may help personalize tapering protocols, minimizing flare risk while optimizing long-term disease management in RA patients.
Methods
The OPTIBIO trial (EudraCT 2012–004482-40) was a phase IV, randomized, open-label, non-inferiority study conducted in five hospitals in Spain. RA patients in sustained remission on stable bDMARD therapy were randomized 1:1 to standard care or dose reduction. The primary outcome was to compare the proportion of joint flare between baseline and 12 months by a non-inferiority analysis analysed by the intention-to-treat principle and to identify predictors for flare and sustained remission.
Results
A total of 195 patients were randomized: 99 to the control group and 96 to the optimization group. Thirty-nine flares occurred (optimization: 22.7%, control: 17.2%), with a risk difference of −5.5% (95% CI: −16.8% to 5.7%; P = 0.33). Two predictive models were developed: one for flares (AUC: 0.84) including 3v-DAS28-CRP, VAS pain, erosions, systolic blood pressure and haemoglobin, and another for sustained remission (AUC: 0.77) including 3v-DAS28-CRP, age and rheumatoid factor. Adding molecular biomarkers improved AUCs to 0.91 and 0.88, respectively. No significant differences in adverse events were observed.
Conclusion
bDMARD dose optimization was not non-inferior to standard therapy on the flare rate but demonstrated similar safety. Predictive models for remission and flares were developed, which may help select patients to ensure safe implementation of this strategy, highlighting the need for personalized treatment.
Trial registration
www.clinicaltrialsregister.eu; EudraCT 2012–004482-40
Keywords: rheumatoid arthritis, bDMARDs, flares, sustained remission, predictive models, personalized medicine
Rheumatology key messages.
Biologic disease-modifying antirheumatic drug (bDMARD) dose optimization was not non-inferior to standard therapy on the flare.
Clinical models classified patients in terms of disease severity.
Incorporation of molecular markers to clinical models improved their predictive capability.
Introduction
The RA treatment aims to achieve remission or maintain low disease activity (LDA) to prevent irreversible joint damage, reduce disability and enhance patients’ quality of life [1]. The American College of Rheumatology (ACR) and the European Alliance of Associations for Rheumatology (EULAR) recommend a treat-to-target (T2T) strategy, which emphasizes regular monitoring of disease activity and therapy adjustments to sustain optimal disease control [2, 3].
The introduction of bDMARD and Janus kinase inhibitors (JAKi) has significantly improved treatment outcomes, enabling more patients to maintain disease remission. This has led to growing interest in the potential for tapering biologic therapies to reduce medication exposure and associated risks while preserving disease control. Observational studies have examined the feasibility of dose reduction, reporting variable adherence to tapered regimens and differing impacts on disease stability [4, 5].
The ACR and EULAR guidelines provide recommendations on bDMARD dose reduction in RA [2, 3]. The US guideline conditionally favours continuing the dose due to low-certainty evidence [3], whereas the European guideline recommends reduction only after glucocorticoid discontinuation and at least 6 months of sustained remission [2]. Despite these recommendations, further data are needed to improve prediction and prevent flares.
Randomized controlled trials (RCTs) have investigated tapering strategies to determine whether reduced dosing can maintain disease control comparable to standard therapy. Findings, however, have been inconsistent across different biologics. Some RCTs reported a higher risk of flares and a reduced likelihood of maintaining remission [6, 7], while others found no statistically significant increase in risk [8, 9]. Factors such as treatment type, patient characteristics and disease history appear to influence the success of dose tapering. Previous studies have shown that patients with early RA tend to respond better to bDMARD dose reduction than those with long-standing disease [10]. Efforts to identify predictors of flare during dose reduction have highlighted several clinical factors linked to increased or decreased relapse risk, including anticitrullinated protein antibodies (ACPA), rheumatoid factor (RF), corticosteroid use, gender and imaging evidence of subclinical inflammation [11–20]. However, no single predictor has proven universally reliable, underscoring the need for a precise approach to guide tapering decisions.
Given the variability in outcomes, further research is needed to refine patient selection criteria and develop strategies that balance efficacy and safety. Identifying robust predictive markers could enable personalized tapering protocols, minimizing flare risk while optimizing long-term disease control in RA patients.
The aim of this study was to compare the proportion of joint flare between baseline and 12 months by a non-inferiority analysis and to identify predictors to optimize treatment and personalize care.
Methods
Patients and methods
The Biologic Optimization Study (OPTIBIO) was a Phase IV, randomized, controlled, open-label, non-inferiority trial conducted in patients with RA in sustained remission on biological therapy. The recruitment occurred at five hospitals in Spain. The study adhered to ethical principles in biomedical research and the current legislation in Spain, following Good Clinical Practice (ICH Topic E6, 2011), the Declaration of Helsinki and the requirements of RD 223/2004. The Autonomous Ethics Committee of Galicia (CEIC) approved the study, and all patients provided written informed consent.
Eligible patients were required to be at least 18 years, RA diagnosis according to the ACR 1987 criteria and have been in remission with stable bDMARD therapy for at least six months prior to screening. Remission was defined using one of the following: a Disease Activity Score 28 (DAS28) ≤2.6 [including C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), or those with three variables], a Simplified Disease Activity Index (SDAI) ≤3.3 or meeting ACR/EULAR Boolean criteria.
Use of conventional synthetic DMARDs (csDMARDs) was permitted, on stable doses for at least six months prior to inclusion. Oral steroids at doses ≤10 mg/day of prednisone (or equivalent) were also eligible. Those who had received intra-articular steroid injections within the past six months were excluded.
Randomization
Patients were randomly assigned in a 1:1 ratio to either optimize their bDMARD therapy or continue with the standard dose. Randomization was centralized using a stratified allocation method based on each patient’s current treatment. Allocation was concealed until the randomization process was completed to ensure impartiality.
Procedures
Patients in the optimization group reduced their bDMARD therapy by extending the dosing interval or by reducing the dose, while the control group continued treatment at the guideline recommended (Supplementary Figure S1). They were followed up with visits every 12 weeks and additional visits scheduled for flares or any events of interest.
Patients who experienced a joint flare in the optimization group returned to their baseline bDMARD dose, while those in the control group received care per current treatment guidelines. No re-optimization was attempted in patients who regained remission after treatment reintroduction.
Outcomes
The primary outcome was to assess the flare-rate between baseline and 12 months through a non-inferiority analysis. Disease activity was evaluated at each centre by an independent, blinded rheumatologist. Patients who maintained a DAS score <2.6, SDAI <3.3 or met ACR/EULAR Boolean remission criteria were defined as sustained remission. A joint flare was defined as the simultaneous occurrence of a DAS28 score >2.6, an SDAI score >3.3 and the loss of ACR/EULAR Boolean remission criteria. The requirement for all three parameters to be present in each patient was intended to ensure a rigorous and precise definition of flare. This approach aimed to capture only clinically meaningful disease activity, thereby minimizing the risk of misclassifying patients. Secondary endpoints were to compare flare-free survival at 18, 24 and 36 months; to develop two prediction models, one for flare and one for sustained remission; and to evaluate the accuracy of each model. Finally, potential clinical and laboratory adverse events were assessed.
Targeted next-generation sequencing
The Ampliseq IAD211967_182 panel was used to sequence 22 selected genes based on extensive curation of peer-reviewed literature and disease research databases (Supplementary Data S2). A total of 195 DNA samples were analysed (96 optimization group, 99 control). Libraries were prepared using Ion Chef and sequenced on an Ion S5XL, obtaining ∼20 million reads per chip with an average depth of 1300×.
Analysis of sequencing data
Reads were aligned to GRCh37 using Torrent Suite. Variants were called with low stringency, filtered by depth, PHRED score and frequency. Aggregated VCF were filtered by MAF <0.01. Plink converted data for logistic regression on flare/remission, including the additive effect of the allelic dose of each variant. Results were annotated using SnpEff.
Anti-cytokine autoantibody (ACAA) analysis
Serum biomarkers were collected at baseline visit. The MILLIPLEX MAP panel was used to detect and quantify 15 anti-cytokine autoantibodies in serum using xMAP technology. Samples were diluted, incubated and analysed on a MagPix instrument (Luminex Corporation; Austin, TX, USA). Assay precision was assessed by calculating the coefficient of variation for replicate samples withing each plate. The complete molecular analysis is found in the Supplementary Data S1-3.
Statistical analyses
A total sample size of 195 patients was calculated. This sample size allowed us to evaluate the non-inferiority hypothesis with a power of 80% and an alpha value of 0.05 with the assumption of a remission rate at one year of 92% in the standard therapy group, a non-inferiority margin of 10% and considering a 10% loss to follow-up.
Baseline characteristics were described by frequencies and percentages for categorical variables and mean (SD) and median (IQR) for continuous variables. The Student’s t test or Mann–Whitney U test was used to compare means between treatment groups, depending on the normality analysis with the Kolmogorov–Smirnov test. The Pearson χ2 test or Fisher’s exact test was applied to assess the relationship between qualitative variables.
For the primary and safety analyses, data were analysed using the intention-to-treat (ITT) population. The proportion of joint flares during the 12-month follow-up was compared between the two groups using the χ2 test. Flare-free survival was assessed using the Kaplan–Meier method, and hazard ratios (HRs) were estimated through a Cox proportional hazards model, using the control group as the reference. Differences between survival curves were evaluated using the log-rank test.
Secondary outcomes were analysed by the same methodological approach used in the primary analysis for the proportion of flares and flare-free survival analyses at 18, 24 and 36 months.
Univariate and multivariate Cox proportional hazards regression analyses were conducted to evaluate the risk of joint flare, HRs and their corresponding 95% CI reported. Additionally, univariate and multivariate logistic regression analyses were performed to assess the odds of achieving sustained remission, with odds ratios (ORs) and 95% CI calculated. For the selection of the final clinical model, the variables with P < 0.20 in the univariate models and those clinically relevant according to the literature were included. Manual elimination of covariates was carried out backwards until the final model was obtained. Subsequently, the complete model was obtained using clinical and molecular variables. Finally, an internal validation of the models was performed using a bootstrap method (500 replicates) to calculate the corrected AUC, with the aim of evaluating the predictive accuracy of the models at 12 months. An AUC exceeding 0.75 was interpreted as indicative of good predictive performance. The optimization group was selected to perform the prediction models. All statistical analyses were conducted using SPSS v28 and R v4.3.1 software. Statistical significance level of P < 0.05 was considered.
This trial was registered in EudraCT with identification number: 2012–004482-40 (www.clinicaltrialsregister.eu).
This manuscript used ChatGPT to improve the clarity and grammar of the language.
Results
Two hundred and forty-one patients were evaluated for eligibility. Forty-six were excluded before randomization: 17 presented a flare, 18 did not meet all inclusion criteria, five for patient decision, five for medical decision and one for missing data. A total of 195 patients were enrolled and randomly assigned: 99 to the control group and 96 to the optimization group. During the first year of follow-up, four patients in the optimization group and three in the control group were lost (Fig. 1).
Figure 1.
Flowchart of RA patients randomized in the clinical trial. IC: informed consent; SAE: Serious Adverse Event; * cutaneous melanoma; ** space occupying lesion of the brain.
Baseline characteristics were comparable between groups (Table 1). The mean age of participants was 63 years (SD 12.78) and 83% were women. The mean remission duration was 22.19 months (SD 16.07) for the control group and 20.82 months (SD 14.55) for the optimization group.
Table 1.
Baseline characteristics of the study population.
| Control group (n = 99) | Optimization group (n = 96) | |
|---|---|---|
| Clinical and demographics | ||
| Age, years | 63.15 (12.22) | 62.84 (13.39) |
| Female | 81.00 (81.82%) | 81.00 (84.38%) |
| Male | 18 (18.18%) | 15 (15.63%) |
| BMI (kg/m2) | 27.54 (4.77) | 27.69 (5.83) |
| Current or previous smoker | 36.00 (36.36%) | 36.00 (37.11%) |
| Never smoking | 63.00 (63.64%) | 61.00 (62.89%) |
| Positive rheumatoid factor (RF) | 71.00 (71.72%) | 67.00 (69.79%) |
| Positive for anti-citrullinated peptide antibodies (ACPA) | 60.00 (65.22%) | 65.00 (68.42%) |
| Double positive (ACPA+RF) | 57.00 (57.58%) | 59.00 (61.46%) |
| Disease duration, months | 172.9 (113.034) | 167.33 (112.70) |
| Remission duration, months | 22.19 (16.07) | 20.82 (14.55) |
| Measure of disease activity | ||
| DAS28 | 1.97 (0.89) | 2.02 (0.76) |
| DAS28-CRP | 1.72 (0.50) | 1.66 (0.46) |
| DAS28-3 | 2.00 (0.90) | 2.09 (0.78) |
| DAS28-3-CRP | 1.69 (0.46) | 1.66 (0.39) |
| SDAI | 3.33 (3.00) | 3.04 (2.99) |
| ESR (normal value <20 mm/h) | 15.51 (16.98) | 15.15 (12.41) |
| CRP (normal value <1mg/dL) | 0.25 (0.38) | 0.20 (0.24) |
| Positive erosion | 58.00 (61.05%) | 61.00 (66.30%) |
| Swollen-joint count 28 ≤ 1 | 96.00 (96.97%) | 96.00 (100%) |
| Tender-joint count 28 ≤ 1 | 85.00 (85.86%) | 84.00 (87.50%) |
| Global and functional assessment | ||
| PhGA (0–10) | 7.79 (8.33) | 7.98 (10.23) |
| PGA (0–100) | 19.28 (19.65) | 16.60 (18.19) |
| VAS for pain (0–100 mm) | 17.07 (18.83) | 13.47 (15.69) |
| HAQ-DI | 0.73 (0.63) | 0.57 (0.62) |
| Treatment | ||
| Adalimumab | 17 (17.17%) | 19 (19.79%) |
| Etanercept | 22 (22.22%) | 23 (23.13%) |
| Infliximab | 5 (5.00%) | 8 (8.30%) |
| Golimumab | 2 (2.00%) | 3 (3.13%) |
| Certolizumab | 11 (11.11%) | 6 (6.19%) |
| Abatacept | 11 (11.11%) | 8 (8.25%) |
| Tocilizumab | 31 (31.31%) | 29 (30.21%) |
| Methotrexate | 29 (30.85%) | 30 (31.91%) |
| Leflunomide | 12 (12.77%) | 9 (9.57%) |
| Sulfasalazine | 2 (2.13%) | 2 (2.13%) |
| Glucocorticoids | 14 (14.40%) | 17 (17.71%) |
Data are n (%), mean (SD) or median (IQR). BMI: body-mass index; CRP: C-reactive protein; DAS28: Disease Activity Score 28; ERS: erythrocyte sedimentation rate; DAS28-3: three variables DAS28; PhGA: physician global assessment; PGA: patient global assessment; HAQ-DI: Health Assessment Questionnaire of Disease Activity; SDAI: Simplified Disease Activity Index.
From baseline to 12 months, 39 flares were recorded: 17 (17.2%) in the control group and 22 (22.7%) in the optimization group, yielding a risk difference of –5.5% (–16.8–5.7; P = 0.33). However, the lower limit of the 95% confidence interval exceeded the non-inferiority margin of –10%. Thus, non-inferiority could not be demonstrated at 12 months (Fig. 2A).
Figure 2.
Non-inferiority analysis and flare-free survival curve. (A) Non-inferiority analysis of the comparison of flare rate (primary objective) and additional analyses. The primary end point was the proportion of relapses in each group using non-inferiority analysis. The dotted vertical red line represents the non-inferiority margin (10%). At 12 months, the 95% confidence interval exceeded the established lower limit; this result was maintained throughout the study. (B) Flare-free survival curve throughout follow-up. The analysis was done in the intention to treat population. Red lines indicate the analyses at each time point. No significant difference was found between the survival of both groups at one year of follow-up. Flare-free survival was higher in the control group and was statistically significant from month 18 of follow-up. The risk of flare was analysed by a Cox Proportional Hazard Regression with 95% CI representing the risk for the optimization group compared with the control group. CI: confidence interval; HR: hazard ratio
Analysis of flare-free survival showed a HR = 1.33 (0.71–2.51) over 12 months of follow-up, which did not reach statistical significance. However, from baseline to 18 and 24 months the flare-free survival was statistically significant (Fig. 2B).
In Univariate Cox regression, older age, higher baseline disease activity scores (DAS28-CRP, 3v-DAS28-CRP and SDAI), tender joint counts, elevated CRP levels, higher Visual Analogue Scale (VAS) pain and greater disability were significantly associated with a risk of flares. On a molecular level, joint flares were associated with seven single nucleotide polymorphisms (SNPs) in genes related to disease activity or treatment response (Supplementary Table S2–3).
In the clinical model, higher disease activity (3v-DAS28-CRP) emerged as the strongest predictor of flare (HR = 14.53 [2.82–74.73]). Elevated haemoglobin levels and the presence of erosions were also associated with increased risk (HR = 1.77 [1.09–2.89] and HR = 7.28 [1.58–33.55], respectively). In addition, both systolic blood pressure (SBP) and VAS pain contributed with modest effects (HR = 1.04 [1.00–1.07] and HR = 1.04 [1.01–1.07], respectively) (Table 2).
Table 2.
Joint flare prediction models.
| HR | (IC, 95%) | P value | |
|---|---|---|---|
| Clinical model | |||
| DAS28-3-CRP | 14.53 | 2.82–74.73 | 0.001 |
| Hb (g/dL) | 1.77 | 1.09–2.89 | 0.020 |
| Erosion | 7.28 | 1.58–33.55 | 0.010 |
| SBP (mm/Hg) | 1.04 | 1.007–1.07 | 0.016 |
| VAS pain | 1.04 | 1.01–1.07 | 0.007 |
| Complete model | |||
| DAS28-3-CRP | 7.28 | 1.43–36.95 | 0.016 |
| Hb (g/dL) | 1.52 | 0.88–2.60 | 0.125 |
| SBP (mm/Hg) | 1.02 | 1.001–1.05 | 0.040 |
| Erosion | 8.15 | 1.58–42.02 | 0.012 |
| VAS pain | 1.06 | 1.02–1.10 | <0.001 |
| rs5746065 (C/A) | 0.31 | 0.09–1.004 | 0.050 |
| rs9594987 (T/C) | 0.15 | 0.04–0.57 | 0.005 |
| Anti-IFNγ | 1.0005 | 1.00004–1.0009 | 0.031 |
DAS28-3-CRP: three variables DAS28; Hb: haemoglobin; SBP: systolic blood pressure; VAS: visual analogue scale.
After adding molecular variables, 3v-DAS28-CRP and erosions remained as strong predictors of flare (HR = 7.28 [1.43–36.95] and HR = 8.15 [1.58–42.02]). Pain intensity retained significance (HR = 1.06 [1.02–1.10]), while SBP showed a smaller effect (HR = 1.02 [1.00–1.05]). Haemoglobin, however, was no longer significant (HR = 1.52 [0.88–2.60]). Among molecular variables, rs5746065 (C/A) in TNFRSF1B gene showed a borderline protective effect (HR = 0.31 [0.09–1.00]), whereas rs9594987 (T/C) in ENOX gene was significantly protective (HR = 0.15 [0.04–0.57]). Anti-IFNγ levels were also linked to increased flare risk (HR = 1.00 [1.00–1.00]) (Table 2). The model demonstrated good discriminative performance at 12 months, with a concordance C-index of 0.884, and was highly significant overall χ2 = 44.18, 8 df; P = 5 × 10−7.
The most accurate clinical model yielded an AUC = 0.84 (0.71–0.97). The inclusion of molecular biomarkers improved the model’s discriminatory capacity, increasing the AUC value to 0.91 (0.82–0.99). This model demonstrated a sensitivity of 75% and specificity of 93%, with negative and positive predictive values (NPV/PPV) of 93% and 76%, respectively (Fig. 3A and B).
Figure 3.
ROC curve of the clinical and complete (clinical+molecular) flare and sustained remission prediction models. (A) Clinical Flare Prediction Model showing an AUC = 0.84 (0.71–0.97). SE of 68%, SP of 93%, PPV of 73% and a NPV of 91%. (B) Complete Flare Prediction Model. AUC = 0.91 (0.82–0.99). SE of 75%, a SP of 93%, a PPV of 76% and a NPV of 93%. The curve represents the relationship between sensitivity and 1-specificity, evaluating the discriminative performance of the model. (C) Clinical Sustained Remission Prediction Model shows AUC = 0.77 [0.66–0.87]. SE of 82%, SP of 63%, PPV of 88% and NPV of 51%. (D) Complete Sustained Remission Prediction Model. AUC = 0.88 (0.78–0.97) SE of 84%, a SP of 83%, a PPV of 95% and a NPV of 60%. The curve represents the relationship between sensitivity and 1-specificity, evaluating the discriminative performance of the model. AUC: area under the curve; SE: sensitivity; SP: specificity; PPV: positive predictive value; NPV: negative predictive value
For sustained remission, univariate analysis showed that age was negatively correlated with remission. In contrast, a higher heart rate was positively associated with it. Pain intensity and the number of painful joints were also negatively correlated. Similarly, greater disease activity (3v-DAS28-PCR) and higher disability scores (HAQ-DI) reduced the likelihood of maintain remission. Additionally, seven SNPs were significantly associated with the probability of remission (Supplementary Table S4–5).
In the multivariate logistic regression model, higher DAS28, remained significantly associated with a reduced likelihood of remission. Specifically, each unit increase in DAS28 corresponded to a 90% decrease in the odds of remission. Age also showed a significant negative effect, with each additional year decreasing the odds of remission by 7%. In contrast, the presence of RF was positively associated with remission, increasing the odds by more than threefold (Table 3).
Table 3.
Sustained remission prediction models in the optimization group.
| OR | (IC 95%) | P value | |
|---|---|---|---|
| Clinical model | |||
| Age (years) | 0.93 | 0.88–0.97 | 0.007 |
| Positive rheumatoid factor | 3.47 | 1.04–12.34 | 0.044 |
| DAS28-3-CRP | 0.10 | 0.01–0.45 | 0.005 |
| Complete model | |||
| Age (years) | 0.93 | 0.88–0.97 | 0.007 |
| Positive rheumatoid factor | 3.47 | 1.04–12.34 | 0.044 |
| DAS28-3-CRP | 0.10 | 0.01–0.45 | 0.005 |
| rs5746065 (C/A) | 5.77 | 1.30–29.38 | 0.024 |
| rs6555900 (G/C) | 2.30 | 0.99–5.83 | 0.059 |
| rs1560011 (G/A) | 3.27 | 1.09–11.73 | 0.046 |
CRP: C-reactive protein; DAS28: Disease Activity Score 28; DAS28-3: three variables DAS28.
The complete model to predict remission at 12 months incorporating both clinical and molecular variables confirmed the significant associations observed in the initial analysis. 3v-DAS28-PCR remained a strong predictor (OR = 0.10 [0.01–0.67]). RF continued to show a positive association, increasing the odds of remission by fourfold (OR = 4.86 [0.98–28.67]), while age retained its negative effect, with each additional year reducing the odds by 8% (OR = 0.92 [0.86–0.97]). Three SNPs were also significantly associated with sustained remission. The rs5746065 (C/A) in TNFRSF1B gene had a strong positive association, with an OR = 5.77 (1.30–29.38), indicating an almost six-fold increase in the likelihood of remission. Similarly, rs6555900 (G/C) in KCNIP gene and rs1560011 (G/A) in CLEC2D gene, with odds of 2.30 (0.99–5.83) and 3.27 (1.09–11.73), respectively (Table 3). The best clinical model showed good discrimination with an AUC = 0.77 (0.66–0.87). Adding molecular biomarkers improved performance, increasing the AUC to 0.88 (0.78–0.97), with 84% sensitivity, 83% specificity and higher predictive values (Fig. 3C and D).
Regarding safety outcomes, no significant differences were observed between the reduced-dose and standard therapy groups (Supplementary Table S6).
Discussion
The OPTIBIO study did not demonstrate non-inferiority of bDMARD dose reduction compared with continued therapy over the 12 months period, and flare rates were significantly higher between baseline and 18 months in the optimization group. Flare-free survival was also superior with standard therapy, suggesting that long-term dose reduction may compromise clinical stability. Safety outcomes were comparable between groups.
The 12-month relapse rate was 22%, prior studies have described flare rates following TNFi dose reduction ranging from 12% to 63%, depending on differences in dose tapering strategies, patient characteristics and definitions of flares used across studies [6, 20]. Our findings are consistent with trials such as ARCTIC REWIND, which also failed to show non-inferiority with dose reduction, though reported a higher flare rate (63%), likely due to a more aggressive tapering protocol [6].
Conversely, some studies suggest that dose reduction may not increase flare risk under specific conditions, indicating a potential for non-inferiority when strict eligibility and monitoring protocols are applied [8, 9]. However, such comparisons remain challenging due to heterogeneity in study design, inclusion criteria and definitions of remission and flare. Importantly, the non-inferiority margin used in our trial was 10%, a more stringent threshold compared with the 20% margin commonly employed in other studies [6].
This conservative approach enhances the clinical relevance and methodological rigor of our findings by setting a higher bar for demonstrating this effect, which is very important to define any predictive tool.
A meta-analysis by Uhrenholt et al. further supported by Henaux et al. concluded that bDMARD or tsDMARD dose reduction is associated with an increased risk of flare or loss of remission [21, 22]. Although these associations did not always reach statistical significance due to heterogeneity, another systematic review corroborated these findings and underscored the importance of careful patient selection in implementing dose optimization strategies [23].
Useful tools have been developed to support bDMARD dose reduction, such as the FLARE-RA questionnaire, whose higher scores correlated with future flares, making it valuable for guiding tapering decisions. Nevertheless, additional tools are still needed to complement existing instruments and further enhance personalized treatment strategies [24].
To better predict flare risk after dose reduction, we developed two models: a clinical model and a complete model (incorporating clinical and molecular data). The clinical model identified traditional predictors such as higher disease activity (3v-DAS28-PCR), greater disability (HAQ-DI) and structural damage, all associated with increased flare risk, even in patients deemed to be in remission. This supports prior literature where markers of residual inflammation, even at subclinical levels, increase the likelihood of relapse [14, 16, 17, 20]. Interestingly, higher haemoglobin levels and higher SBP were also associated with an increased risk of flare in our study. This finding regarding haemoglobin appears to contrast with previous reports, where lower haemoglobin levels have been associated with higher disease activity, reflecting the well-established link between chronic inflammation and anaemia. Regarding SBP, our findings are supported by evidence linking chronic systemic inflammation with increased cardiovascular risk, including the development of hypertension [25].
The complete model added two SNPs and anti-IFN-γ levels, improving its predictive capacity. Elevated anti-IFN-γ levels emerged as a potential biomarker for flares, suggesting new avenues for biomarker discovery [12, 26].
One of the identified SNPs, rs9594987 in the ENOX1 gene, has been previously associated with RA severity and potential response to tocilizumab, although the evidence remains conflicting [27]. The inclusion of such genetic markers may improve future efforts in personalized medicine and flare prevention.
Parallel to the flare models, we developed two models to predict sustained remission following dose reduction. In the clinical model, younger age, RF seropositivity and lower baseline 3v-DAS28-PCR scores were all associated with a higher likelihood of maintaining remission. These findings underscore the relevance of achieving deep remission before attempting dose tapering and support the prognostic value of serological markers in RA. Interestingly, contrary to previous studies [11, 14, 17, 28] variables such as disease duration, time to remission, ACPA/RF seropositivity and gender were not significantly associated with remission in our model. This highlights the variability and limitations of conventional clinical predictors across different cohorts.
The complete model for sustained remission incorporated three SNPs, improving the model’s accuracy. Among these, rs1560011 in the CLEC2D gene has been previously studied in relation to tocilizumab response and low disease activity [29]. A novel finding was the rs6555900 variant in the KCNIP1 gene, which encodes a potassium channel regulator, not previously described in RA. Additionally, the rs5746065 variant in the TNFRSF1B gene, encoding the TNF-α receptor 2 (TNFR2), was associated with a double protective effect: reduced flare risk and increased likelihood of sustained remission [30].
These results support the potential utility of molecular data in identifying patients who may safely undergo dose tapering, particularly when conventional clinical markers are insufficient.
This study has several strengths. First, its randomized controlled trial (RCT) design enhanced internal validity and supported the robustness of the primary outcome findings. Second, the extended follow-up period of up to 36 months allowed the assessment of both early and delayed flares, providing a comprehensive understanding of the long-term impact of dose reduction. Importantly, the study applied a stringent definition of flare and remission, using three validated instruments (DAS28-CRP, Boolean remission criteria and SDAI) and a tighter 10% non-inferiority margin than in previous studies. These rigorous criteria ensured precise event classification and provided a solid foundation for developing a reliable and robust predictive tool for flare and sustained remission.
Limitations should be acknowledged. The open-label design may have introduced performance or expectation bias, partially mitigated by using blinded rheumatologists for disease activity assessment. The absence of imaging data could have missed subclinical inflammation preceding flares, affecting remission interpretation. The 1987 ACR criteria may underrepresent early or seronegative RA, although anti-CCP positive patients were included. While DAS28-3V was the strongest predictor, its clinical interpretation is limited due to lack of validated MCID cut-offs. Finally, the genetic analyses were performed solely in a Caucasian population, limiting generalizability and underscoring the need for validation in diverse cohorts.
In conclusion, bDMARD dose reduction was not non-inferior to continued therapy on the flare rate, although safety profiles were comparable. The development of accurate clinical and molecular models to predict both flare risk and sustained remission represents an important step toward precision medicine in RA. These tools may guide personalized treatment decisions, allowing safe implementation of tapering strategies in appropriately selected patients.
Supplementary Material
Contributor Information
Francisco J Blanco, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Laura Galindo, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Belen Acasuso, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Vanesa Balboa-Barreiro, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Juan D Cañete, Department of Rheumatology, Arthritis Unit, Hospital Clinic Barcelona IDIBAPS, Barcelona, Spain.
Benjamin Fernández-Gutiérrez, Department of Rheumatology and Health Research Institute (IdISSC), Hospital Clínico San Carlos, Madrid, Spain.
Isidoro González-Álvaro, Department of Rheumatology, Hospital Universitario La Princesa, IIS, Madrid, Spain.
José Luis Pablos Álvarez, Department of Rheumatology, Instituto de Investigación Hospital 12 de Octubre, Complutense University of Madrid, Madrid, Spain.
Carmen Bejerano-Herrería, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Maite Silva-Díaz, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Ignacio Rego-Perez, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Lucia Lourido, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Cristina Ruiz-Romero, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Miren Uriarte-Ecenarro, Department of Rheumatology, Hospital Universitario La Princesa, IIS, Madrid, Spain.
Rosario García-Vicuña, Department of Rheumatology, Hospital Universitario La Princesa, IIS, Madrid, Spain.
Andrea Cuervo, Department of Rheumatology, Arthritis Unit, Hospital Clinic Barcelona IDIBAPS, Barcelona, Spain.
Julio Ramírez, Department of Rheumatology, Arthritis Unit, Hospital Clinic Barcelona IDIBAPS, Barcelona, Spain.
Raquel Celis, Department of Rheumatology, Arthritis Unit, Hospital Clinic Barcelona IDIBAPS, Barcelona, Spain.
Luis Rodríguez-Rodríguez, Department of Rheumatology and Health Research Institute (IdISSC), Hospital Clínico San Carlos, Madrid, Spain.
Lydia Abasolo Alcázar, Department of Rheumatology and Health Research Institute (IdISSC), Hospital Clínico San Carlos, Madrid, Spain.
Dalifer Freites Nuñez, Department of Rheumatology and Health Research Institute (IdISSC), Hospital Clínico San Carlos, Madrid, Spain.
Maria Martín-López, Department of Rheumatology, Instituto de Investigación Hospital 12 de Octubre, Complutense University of Madrid, Madrid, Spain.
Francisco J Toro-Santos, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Natividad Oreiro, Department of Rheumatology, Rheumatology Research Group (GIR) Biomedical Research Institute (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), Sergas, A Coruña, Spain.
Supplementary material
Supplementary material is available at Rheumatology online.
Data availability
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
Funding
This study has been funded by Instituto de Salud Carlos III (ISCIII) through the projects PMP15/00032 and PMP22/00101 co-funded by the European Regional Development Fund ‘A way to make Europe’; funded by ISCIII, co-funded by the European Union and projects PI20/00793, PI22/01155, PI23/00818 and RD21/0002/0009 funded by Instituto de Salud Carlos III–European Union-NextGenerationEU and supported by CIBER-Consorcio Centro de Investigacion Biomedica en Red- (CB06/01/0040), ISCIII. We also counted upon funding from grant IN607A2021/07 and IN607A_2025/11 from Xunta de Galicia. L.G. is a PhD student supported by Cátedra FSR-UDC, Universidad de A Coruña.
Disclosure statement: F.J.B. has received grants or contracts from speaker fees from Abbvie, Bristol Myers Squibb, Roche, Servieer, Novartis, Horizon Therapeutics Ireland DAC, ITF Research Pharma S.L.U., GSK Research, Pfizer, Sanofi-Aventis, Grunenthal, Lilly, Merck Healthcare, KgaA, LG Chem, Ltd, UCB, Janssen, Amgen, Regeeron, Alkem Labortories Ltd, Grunenthal, Sun Pharma Global FZE, Kiniksa Pharmaceuticals, GmbH, speakers fees from Medicamenta-Ecuador, Grunenthal and Asofarma, support attending meetings from UCB, Abbie and Celgen. I.G.-A. has received grants or contracts from Grebro Pharma and European Union also payments or honoraria for presentations from Lilly and support for attending meetings from Lilly and Pfizer. R.G.-V. has received grants from Abbvie, Lilly, Sanofi and MSD, consulting fees from Biogen, Galápagos, MSD, Sanofi and Johnson&Johnson, payment honoraria for presentations from Pfizer and Abbvie, support for attending meetings from Abbvie, Lly, MSD, Pfizer, Sandoz and UCB and receipt of equipment for her institution from Abbvie (all of them unrelated to the study presented in this manuscript). The remaining authors have declared no conflicts of interest.
References
- 1. Aletaha D, Smolen JS. Diagnosis and management of rheumatoid arthritis: a review. JAMA 2018;320:1360–72. [DOI] [PubMed] [Google Scholar]
- 2. Smolen JS, Landewé RBM, Bergstra SA et al. EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs: 2022 update. Ann Rheum Dis 2023;82:3–18. [DOI] [PubMed] [Google Scholar]
- 3. Fraenkel L, Bathon JM, England BR et al. 2021 American college of rheumatology guideline for the treatment of rheumatoid arthritis. Arthritis Rheumatol (Hoboken, NJ) 2021;73:1108–23. [DOI] [PubMed] [Google Scholar]
- 4. Brahe CH, Krabbe S, Østergaard M et al. Dose tapering and discontinuation of biological therapy in rheumatoid arthritis patients in routine care – 2-year outcomes and predictors. Rheumatol Oxf Engl 2019;58:110–9. [DOI] [PubMed] [Google Scholar]
- 5. Dierckx S, Sokolova T, Lauwerys BR et al. Tapering of biological antirheumatic drugs in rheumatoid arthritis patients is achievable and cost-effective in daily clinical practice: data from the Brussels UCLouvain RA Cohort. Arthritis Res Ther 2020;22:96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Lillegraven S, Paulshus Sundlisæter N, Aga A-B et al. Effect of tapered versus stable treatment with tumour necrosis factor inhibitors on disease flares in patients with rheumatoid arthritis in remission: a randomised, open label, non-inferiority trial. Ann Rheum Dis 2023;82:1394–403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Fautrel B, Pham T, Alfaiate T et al. Step-down strategy of spacing TNF-blocker injections for established rheumatoid arthritis in remission: results of the multicentre non-inferiority randomised open-label controlled trial (STRASS: spacing of TNF-blocker injections in rheumatoid arthritis study). Ann Rheum Dis 2016;75:59–67. [DOI] [PubMed] [Google Scholar]
- 8. Herwaarden N van, Maas A van der, Minten MJM, Hoogen FHJ van den, Kievit W, Vollenhoven RF van, et al. Disease activity guided dose reduction and withdrawal of adalimumab or etanercept compared with usual care in rheumatoid arthritis: open label, randomised controlled, non-inferiority trial. BMJ. 2015;350:h1389. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Bouman CA, van Herwaarden N, van den Hoogen FH et al. Long-term outcomes after disease activity-guided dose reduction of TNF inhibition in rheumatoid arthritis: 3-year data of the DRESS study–a randomised controlled pragmatic non-inferiority strategy trial. Ann Rheum Dis 2017;76:1716–22. [DOI] [PubMed] [Google Scholar]
- 10. Westhovens R, , RoblesM, , Ximenes ACet al. Maintenance of remission following 2 years of standard treatment then dose reduction with abatacept in patients with early rheumatoid arthritis and poor prognosis. Ann Rheum Dis 2015;74:564–8. doi: 10.1136/annrheumdis-2014-206149 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Haschka J, Englbrecht M, Hueber AJ et al. Relapse rates in patients with rheumatoid arthritis in stable remission tapering or stopping antirheumatic therapy: interim results from the prospective randomised controlled RETRO study. Ann Rheum Dis 2016;75:45–51. [DOI] [PubMed] [Google Scholar]
- 12. Kameda H, Hirata A, Katagiri T et al. Prediction of disease flare by biomarkers after discontinuing biologics in patients with rheumatoid arthritis achieving stringent remission. Sci Rep 2021;11:6865. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Vittecoq O, Desouches S, Kozyreff M et al. Relapse in rheumatoid arthritis patients undergoing dose reduction and withdrawal of biologics: are predictable factors more relevant than predictive parameters? An observational prospective real-life study. BMJ Open 2019;9:e031467. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Ghiti Moghadam M, Vonkeman HE, ten Klooster PM, et al. ; on behalf of the Dutch National POET Collaboration. Stopping Tumor necrosis factor inhibitor treatment in patients with established rheumatoid arthritis in remission or with stable low disease activity: a pragmatic multicenter, open-label randomized controlled trial. Arthritis Rheumatol (Hoboken, NJ) 2016;68:1810–7. [DOI] [PubMed] [Google Scholar]
- 15. Ahmad HA, Baker JF, Conaghan PG et al. Prediction of flare following remission and treatment withdrawal in early rheumatoid arthritis: post hoc analysis of a phase IIIb trial with abatacept. Arthritis Res Ther 2022;24:47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. van der Leeuw MS, Messelink MA, Tekstra J et al. Using real-world data to dynamically predict flares during tapering of biological DMARDs in rheumatoid arthritis: development, validation, and potential impact of prediction-aided decisions. Arthritis Res Ther 2022;24:74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Chaiamnuay S, Jiemjit S, Songdechaphipat W et al. Predictors of flare in rheumatoid arthritis patients with persistent clinical remission/low disease activity: data from the TARAC cohort. Medicine (Baltimore) 2022;101:e29974. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Schlager L, Loiskandl M, Aletaha D, Radner H. Predictors of successful discontinuation of biologic and targeted synthetic DMARDs in patients with rheumatoid arthritis in remission or low disease activity: a systematic literature review. Rheumatol Oxf Engl 2020;59:324–34. [DOI] [PubMed] [Google Scholar]
- 19. Emery P, Burmester GR, Naredo E et al. Adalimumab dose tapering in patients with rheumatoid arthritis who are in long-standing clinical remission: results of the phase IV PREDICTRA study. Ann Rheum Dis 2020;79:1023–30. doi: 10.1136/annrheumdis-2020-217246. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Ibrahim F, Lorente-Cánovas B, Doré CJ et al. Optimizing treatment with tumour necrosis factor inhibitors in rheumatoid arthritis—a proof of principle and exploratory trial: is dose tapering practical in good responders? Rheumatology (Oxford) 2017;56:2004–14. doi: 10.1093/rheumatology/kex315. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Uhrenholt L, Christensen R, Dinesen WKH et al. Risk of flare after tapering or withdrawal of biologic/targeted synthetic disease-modifying anti-rheumatic drugs in patients with rheumatoid arthritis or axial spondyloarthritis: a systematic review and meta-analysis. Rheumatology 2022;61:3107–22. [DOI] [PubMed] [Google Scholar]
- 22. Henaux S, Ruyssen-Witrand A, Cantagrel A et al. Risk of losing remission, low disease activity or radiographic progression in case of bDMARD discontinuation or tapering in rheumatoid arthritis: systematic analysis of the literature and meta-analysis. Ann Rheum Dis 2018;77:515–22. doi: 10.1136/annrheumdis-2017-212423. [DOI] [PubMed] [Google Scholar]
- 23.Rotea S, Galindo L, Acasuso B et al. Efficacy of tapering biologics and JAK inhibitors in rheumatoid arthritis: A systematic review and meta-analysis. [under review]. Complejo Hospitalario Universitario de la Coruña. doi: 10.1101/2025.09.29.25336875 [DOI] [Google Scholar]
- 24. Doumen M, , BertrandD, , Pazmino Set al. Validation and predictive capacity of a Dutch version of the FLARE-RA questionnaire within the context of a TNFi-tapering trial. Clin Rheumatol 2023;42:39–45. [DOI] [PubMed] [Google Scholar]
- 25. Allaart CF, Goekoop-Ruiterman YPM, de Vries-Bouwstra JK, Breedveld FC, Dijkmans BAC; FARR study group. Aiming at low disease activity in rheumatoid arthritis with initial combination therapy or initial monotherapy strategies: the BeSt study. Clin Exp Rheumatol 2006;24:S77–82. [PubMed] [Google Scholar]
- 26. Sakashita E, Nagatani K, Endo H, Minota S. Serum level of IFNβ distinguishes early from late relapses after biologics withdrawal in rheumatoid arthritis. Sci Rep 2022;12:16547. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Mikhaylenko DS, Nemtsova MV, Bure IV et al. Genetic polymorphisms associated with rheumatoid arthritis development and antirheumatic therapy response. Int J Mol Sci 2020;21:4911. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Kedra J, Dieudé P, Giboin C, et al. ; CRI‐IMIDIATE clinical research network. Towards the lowest efficacious dose: results from a multicenter noninferiority randomized open-label controlled trial assessing tocilizumab or abatacept injection spacing in rheumatoid arthritis in remission. Arthritis Rheumatol 2024;76:541–52. doi: 10.1002/art.42752. [DOI] [PubMed] [Google Scholar]
- 29. Maldonado-Montoro M, Cañadas-Garre M, González-Utrilla A, Plaza-Plaza JC, Calleja-Hernández MŸ. Genetic and clinical biomarkers of tocilizumab response in patients with rheumatoid arthritis. Pharmacol Res 2016;111:264–71. [DOI] [PubMed] [Google Scholar]
- 30. Nouri B, Nair N, Barton A. Predicting treatment response to IL6R blockers in rheumatoid arthritis. Rheumatology (Oxford) 2020;59:3603–10. doi: 10.1093/rheumatology/keaa529. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.