FormalPara Key Summary Points

Why carry out this study?

Efficacy and benefit/risk of baricitinib have been demonstrated in randomised controlled trials but large, multinational, real-world evidence is limited.

What was learned from the study?

Real-world evidence regarding time to discontinuation, effectiveness, treatment patterns, as well as clinical and patient-reported outcomes in patients with RA initiating treatment with baricitinib, any other targeted synthetic disease-modifying anti-rheumatic drug (tsDMARD), or any biologic disease-modifying anti-rheumatic drug (bDMARD) in clinical practice.

Evidence from real-world studies can better inform clinicians treating patients in clinical practice.

Although patients initiating baricitinib were older with a longer disease duration, time to discontinuation was greater for baricitinib compared to other treatments and effectiveness outcomes were consistent regardless of prior treatment or age.

Introduction

The aim of the current treat-to-target strategy for rheumatoid arthritis (RA) is to achieve sustained disease remission, or at the least low disease activity (LDA), through regular monitoring of disease activity every three months and adaption of therapy if remission or LDA is not reached [1]. Treating to achieve remission or LDA may reduce long-term joint damage and may improve patient outcomes [2]. To achieve the treat-to-target goal, conventional synthetic disease-modifying anti-rheumatic drugs (csDMARDs; predominately methotrexate), biologic disease-modifying anti-rheumatic drugs (bDMARDs; e.g. tumour necrosis factor inhibitors [TNFi], interleukin-6 inhibitors, anti-CD20 antibody, cytotoxic T-lymphocyte-associated protein 4 [CTLA-4] agonists), or targeted synthetic disease-modifying anti-rheumatic drugs (tsDMARDs; e.g. Janus kinase [JAK] inhibitors) are typically prescribed.

Baricitinib is a preferentially selective and reversible inhibitor of JAK1 and JAK2 approved as a monotherapy and in combination with any csDMARD for the treatment of adults with moderate-to-severe active RA, moderate-to-severe atopic dermatitis, and severe alopecia areata, in Europe, Japan, and multiple other countries. Baricitinib has demonstrated superior efficacy compared to a TNFi in patients with inadequate response to methotrexate (RA-BEAM) [3]. Several randomised controlled trials have reported the efficacy and safety of baricitinib across different patient populations with RA, including naïve (RA-BEGIN), inadequate responders to methotrexate (RA-BEAM), and csDMARDs (RA-BUILD), as well as inadequate responders to bDMARDs (RA-BEACON) [4,5,6]. In addition, greate and more rapid pain relief has been reported with baricitinib treatment, with a postulated direct effect independent of inflammation control [7, 8]. Long-term extension studies (RA-BEYOND) have demonstrated efficacy up to seven years and relative safety up to 9.3 years with baricitinib treatment in patients with RA [9]. Furthermore, long-term relative safety of baricitinib has been reported across multiple development programmes including atopic dermatitis, alopecia areata, systemic lupus erythematosus, and COVID-19 [10].

Clinical studies have provided comprehensive efficacy and safety data; however, patients with RA in real-world clinical settings may typically be older, have a longer disease duration, and greater DMARD experience or comorbidities compared to those in a clinical trial setting [11]. As such, real-world evidence is helpful to fully capture the effectiveness and safety of baricitinib in a population with RA that would not typically fulfil entry criteria for randomised controlled trials, which in turn can better inform clinicians treating patients in everyday practice.

RA-BE-REAL is a 3-year, multinational, prospective, observational study of adult patients with RA. Here we report descriptive baseline characteristics, time to discontinuation, effectiveness, and patient-reported outcomes as well as comparative analyses of time to discontinuation, effectiveness, rates of remission or LDA, and patient-reported outcomes in patients with RA enrolled in Europe following 24 months of either baricitinib, any other tsDMARD or bDMARD, and who initiated that treatment for the first time. Additionally, exploratory analysis in b/tsDMARD-naïve and experienced patients, as well as in patients < 65 and ≥ 65 years old was performed.

Methods

Study Design and Patient Population

RA-BE-REAL is being conducted across five European countries (France, Germany, Italy, Spain, and the UK) as well as Australia, Canada, and Saudi Arabia. The study commenced in October 2018, and completion is expected in October 2024. Patient enrolment for European countries was completed in March 2020, while enrolment is ongoing in other countries. Thus, this predefined interim analysis has been performed on the European subpopulation after 24 months of follow-up.

Eligible patients were aged 18 years or older and met the criteria for RA according to their treating physician and were prescribed baricitinib, any other tsDMARD, or bDMARD per local label requirements for the first time at any point in the treatment algorithm.

The RA-BE-REAL study design has been previously published (Fig.  S1 in the supplementary material) [12]. Patients in cohort A initiated treatment with baricitinib (2 mg or 4 mg), while patients in cohort B initiated any bDMARD or any other tsDMARD (List S1) for the first time. In both cohorts, the targeted DMARD could be used whatever the position in therapeutic lines, but patients had to be naïve to the study molecule. Treatment initiation with baricitinib, any other tsDMARD, or any bDMARD, and treatment changes during the observation period were solely at the discretion of the physician and the patient, in line with locally applicable guidelines and clinical routine. Data were collected at baseline (defined as initiation of treatment with baricitinib, any bDMARD, or any other tsDMARD) and at routine clinical care visits post-baseline at approximately 3, 6, 12, 18, 24, and 36 months. This report describes the results at 24 months (Table 1).

Table 1 Summary of analyses
Full size table

Study Objectives and Endpoints

The primary objective was to assess the time to discontinuation of initial baricitinib, any other tsDMARD, or any bDMARD treatment for all causes (excluding sustained clinical response) over 24 months in this European population.

Comparative effectiveness analyses, over 24 months, included time to treatment discontinuation for all causes (excluding sustained clinical response), percentage of patients achieving Clinical Disease Activity Index (CDAI) remission (defined as CDAI score ≤ 2.8) or LDA (defined as CDAI score > 2.8 and ≤ 10), as well as mean changes from baseline for CDAI, pain visual analogue scale (VAS), and the Health Assessment Questionnaire-Disability Index (HAQ-DI). Only patients with a CDAI score > 2.8 at baseline were included. Adverse events were not actively collected over the duration of this study, only reasons for discontinuation related to safety were collected. Investigators followed their country-specific pharmacovigilance established reporting systems. As a result of the small sample sizes, non-TNFi (biologics excluding TNFi) and tsDMARD groups were pooled as other mechanism of action (OMA). Following similar approaches previously published, patients receiving either baricitinib dosing regimen were pooled [13].

The analyses were conducted in the overall population, as well as after stratification of patients in two ways: (1) b/tsDMARD-naïve, defined as patients who never received any b/tsDMARD (1st line of b/tsDMARD), or b/tsDMARD-experienced, defined as patients who previously received any b/tsDMARD; (2) patients < 65 and ≥ 65 years of age. The ≥ 65 age group was selected on the basis of the recent update to Pharmacovigilance Risk Assessment Committee (PRAC) guidelines [14]. Other risk factors have not been included as only limited risk factors were collected during the study.

Statistical Analyses

For the primary objective, Kaplan–Meier analysis was used to describe time to discontinuation in cohort A and cohort B (including subgroups of cohort B including TNFi, non-TNFi [biologics excluding TNFi], and any other tsDMARD) following 24 months of follow-up.

The comparative analyses were conducted using a frequentist model averaging (FMA) [15] framework based on a data-driven machine learning causal inference approach, which has been previously used to describe comparative effectiveness in real-world settings [16]. This approach reported both the average and best models that were selected from a set of analysis strategies. The analysis strategies were prespecified and derived from a combination of treatment models (used to estimate treatment selection) as well as outcome models (used to estimate the treatment effect). The number and type of strategies varied depending on the outcome of interest. The list of analysis strategies used for the time-to-event as well as the binary and continuous outcomes are listed in Supplemental Table S1. Balancing scores [17] (e.g. propensity scores) were computed using logistic regression and/or tree-based methods (Fig. S2). Variables used in the balancing scores included demographics: age (years), body mass index (BMI; kg/m2), sex, race, education level, current smokers; disease characteristics: family history of RA and RA duration; prior treatment with csDMARDs only or prior treatment with bDMARDs/tsDMARDs (split by TNFi only, other non-TNFi bDMARDs, prior tsDMARDs, all), and concomitant use of csDMARDs at baseline; baseline clinical characteristics: CDAI, HAQ-DI, and pain (VAS; Table S2). To assess whether balance was achieved, standardised difference (acceptable ranges are < 0.25) and variance ratio (acceptable ranges 0.5–2.0) statistics were calculated, graphed, and assessed [18]. Figure S3 and Table S3 display the standardised mean and variance ratio information for the covariates used in the analysis. The propensity score, extreme gradient boosting average standardised absolute mean difference, used for evaluation was derived from the best model taken from the time to discontinuation analysis comparing cohort A and cohort B. All statistics are within the acceptable ranges for balance assessment.

For missing data, outcomes were imputed using non-responder imputation or modified Baseline Observation Carried Forward (Supplement S1) and, as sensitivity analyses, were also reported as observed. For the propensity score logistic regression models and selected outcome models, missing categorical covariates were imputed with the mode and continuous covariates with the mean, missing data were ≤ 10%. For gradient boosted tree models, the treatment model imputation was implemented via the TWANG and XGBoost package in R. Comparative adjusted results are presented as odds ratios (ORs; percentage of patients achieving LDA and remission), least squares mean (LSMean) differences (mean change from baseline [CFB] to endpoint in CDAI, HAQ-DI, and pain [VAS]), or restricted hazard ratios (rHR; time from treatment initiation until treatment discontinuation for all causes) with 95% confidence intervals (CI) formed using the 2.5th and 97.5th percentiles derived from the 100 bootstrap samples. The E-value was used to assess the impact of unmeasured confounding [19]. All analyses were conducted using SAS Version 9.4 (SAS Institute, Cary, NC, USA) and R Version 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria [20]).

Role of Funding Source

The study was funded by Eli Lilly and Company. The study sponsor was involved in the study design, collection, analysis, and interpretation of data, in the writing of this article, as well as the decision to submit this article for publication.

Ethical Approval

This study was conducted in accordance with Good Clinical Practice, the principles of the Declaration of Helsinki, and local laws and regulations in the five European countries. All patients provided written informed consent prior to entry to the study. Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Results

Patient Baseline Characteristics

Of 1073 patients enrolled from France (N = 220), Germany (N = 422), Italy (N = 191), Spain (N = 80), and the UK (N = 160), 510 initiated treatment with baricitinib (cohort A) and 563 initiated treatment with a bDMARD or any other tsDMARD (cohort B).

Baseline demographics and clinical characteristics, as well as RA treatment history, are described in Table 2. Compared with cohort B, patients in cohort A more frequently initiated therapy with baricitinib as a monotherapy. Within the baricitinib group, patients receiving 2 mg dosage were older (72.6 and 57.4 years for 2 mg and 4 mg, respectively), had a longer disease duration (12.3 and 9.8 years for 2 mg and 4 mg, respectively), and a higher proportion were receiving monotherapy (61.0% and 49.2% for 2 mg and 4 mg, respectively) compared with patients receiving 4 mg dosage. Compared with all other treatment groups, patients initiating TNFi had shorter mean disease duration (7.1 years), and a greater proportion were naïve to previous b/tsDMARDs (75.1%).

Table 2 Baseline demographics and clinical characteristics of patients in cohort A and cohort B
Full size table

Post hoc analyses of patients < 65 and ≥ 65 years old demonstrated largely similar BMI, RA treatment patterns, and family history of RA (Table S4). Patients in the < 65 years population had a shorter mean disease duration for both cohorts (Table S4). As expected, the distribution of treatments was similar across age groups. A higher proportion of patients in the TNFi group for both age populations were naïve to b/tsDMARDs. Clinical characteristics and disease activity were similar for all cohorts.

Treatment Discontinuation of Initial RA Therapy

At 24 months, 38.2% (n = 195, median time to discontinuation = not calculable, the median was unable to be calculated as it does not reach the probability of 0.5) of patients treated with baricitinib and 59.9% (n = 337) of patients treated with  a b/tsDMARD had discontinued RA treatment (Fig. 1). Among patients in cohort B, 59.1% (n = 199, median time to discontinuation = 15.6 months, 95% CI [12.2, 20.5]), 57.8% (n = 93, median time to discontinuation = 13.1, 95% CI [8.5, 24.0]), and 69.2% (n = 45, median time to discontinuation = 9.6, 95% CI [6.2, 17.0]) using a TNFi, non-TNFi, and any other tsDMARD, respectively, discontinued treatment at 24 months. Drug survival rates were consistent across overall, b/tsDMARD-naïve, and b/tsDMARD-experienced (Fig. 1), < 65, and ≥ 65-year-old patient subgroups (Fig. S4). Drug survival rates for cohort A were consistently higher than cohort B, regardless of baricitinib dosage or treatment pattern (Fig. S5).

Fig. 1
figure 1

Drug survival over 24 months for overall (a), b/tsDMARD-naïve (b), and b/tsDMARD-experienced patients (c). Instances where the median value is absent, indicates the median was unable to be calculated as it does not reach the probability of 0.5. Drug survival curves for patients < 65 and ≥ 65 years old can be found in Fig. S4. TNFi tumour necrosis factor inhibitor, tsDMARD targeted synthetic disease-modifying anti-rheumatic drugs

Full size image

The comparative adjusted analysis for the primary objective, time to discontinuation, showed a lower discontinuation rate for baricitinib compared to TNFi and other mechanism of action (OMA) in the overall group (baricitinib vs. cohort B overall: rHR 0.6, 95% CI [0.5, 0.7]; vs. TNFi: rHR 0.5, 95% CI [0.4, 0.6]; vs. OMA: rHR 0.6, 95% CI [0.5, 0.7]), as well as naïve and experienced (Fig. 2).

Fig. 2
figure 2

Comparative adjusted analysis using FMA approach for time to discontinuation of baricitinib versus cohort B overall, TNFi, and OMA at 24 months for overall, b/tsDMARD-naïve, b/tsDMARD-experienced, < 65 years, and ≥ 65 years. Results are statistically significant if 1 is not covered by the 95% CI for the restricted hazard ratios. Results of the best performing model can be found in Fig. S6. bDMARD biologic disease-modifying anti-rheumatic drug, CI confidence interval, FMA frequentist model averaging, NRI non-responder imputation, OMA other mechanism of action, TNFi tumour necrosis factor inhibitor, tsDMARD targeted synthetic disease-modifying anti-rheumatic drug

Full size image

Drug survival rates were consistently higher among the baricitinib cohort, compared with the b/tsDMARD cohort for both age cohorts (Fig. S4a, b). The comparative adjusted analysis for time to discontinuation also showed a lower discontinuation rate of baricitinib compared to TNFi and OMA for both age populations (Fig. 2).

In the overall population, discontinuation due to adverse events was similar across all treatment groups, with the greatest proportion of discontinuations in the any other tsDMARD group in cohort B (20.0%; Table 3). Discontinuation due to ineffectiveness was highest for patients discontinuing TNFi (22.6%) and lowest for those discontinuing baricitinib (14.3%). Additionally, a high proportion of patients in cohort B had discontinued treatment at 24 months for other reasons including patient or physician decision, non-compliance, cannot afford medication, unknown causes, COVID-19-related causes (medical and non-medical reasons), or other (TNFi, 29.1%; non-TNFi, 29.2%; tsDMARD, 30.8%) compared to baricitinib (16.1%). Discontinuation due to ineffectiveness was greater among experienced patients (baricitinib, 20.4%; TNFi, 26.2%; non-TNFi, 22.3%; tsDMARD, 29.3%) compared to naïve (baricitinib, 7.8%; TNFi, 21.3%; non-TNFi, 16.4%; tsDMARD, 0%), while discontinuation due to adverse events was similar between both groups (naïve baricitinib, 8.2%; TNFi, 8.3%; non-TNFi, 9.0%; tsDMARD, 16.7%; experienced baricitinib, 7.5%; TNFi, 4.8%; non-TNFi, 8.5%; tsDMARD, 22.0%; Table 3). The reasons for treatment discontinuation were largely consistent between the cohorts for both age groups (Table 3). Discontinuation due to ineffectiveness was lowest among patients treated with baricitinib for both age populations.

Table 3 Reasons for discontinuation in cohort A and cohort B at 24 months
Full size table

Effectiveness

LDA and Remission

For unadjusted results at 24 months, 41.1% of patients overall treated with baricitinib and 36.4% (TNFi), 30.4% (non-TNFi), and 35.5% (tsDMARD) of patients in cohort B achieved LDA (Table 4). Remission was achieved by 15.2% of patients treated with baricitinib and by 16.2% (TNFi), 7.6% (non-TNFi), and 21.0% (tsDMARD) in cohort B.

Table 4 Mean CFB in clinical outcomes and proportion of patients achieving CDAI remission or low disease activity for overall, b/tsDMARD-naïve, b/tsDMARD-experienced, < 65-year-old, and ≥ 65-year-old patients at 24 months
Full size table

Among those naïve to b/tsDMARD therapy, LDA was achieved by 46.6% of patients treated with baricitinib, and 39.0% (TNFi), 46.2% (non-TNFi), and 39.1% (tsDMARD) of patients in cohort B. Among naïve patients treated with baricitinib, 18.1% achieved remission, while 17.9% receiving TNFi, 12.3% receiving non-TNFi, and 26.1% receiving tsDMARDs in cohort B achieved remission. For those with previous b/tsDMARD experience, LDA was achieved by 36.2% of patients treated with baricitinib, and 28.4% (TNFi), 19.4% (non-TNFi), and 33.3% (tsDMARD) of patients in cohort B. Remission was achieved by 12.7% of patients treated with baricitinib, while 11.1% (TNFi), 4.3% (non-TNFi), and 17.9% (tsDMARD) of patients in cohort B achieved remission. Compared to patients experienced to b/tsDMARD therapy, naïve patients achieved an overall higher rate of LDA and remission (Table 4).

For the comparative analysis, LDA response was statistically significant between cohorts A and B across overall (OR 1.5; 95% CI [1.1, 1.9]) and within experienced (OR 1.9; 95% CI [1.2, 2.8]) patients using the imputed data. Additionally, using NRI (non-responder imputation), LDA was statistically significant for baricitinib vs. OMA (OR 1.9; 95% CI [1.2, 3.1]) and vs. TNFi (OR 1.7; 95% CI [1.0, 3.1]) for experienced patients (Fig. 3). There were no statistical differences in remission rates over the 24 months, for both observed and NRI.

Fig. 3
figure 3

Comparative adjusted analysis using FMA approach of remission and LDA for baricitinib versus cohort B overall, TNFi, and OMA at 24 months for overall (a), b/tsDMARD-naïve (b), and b/tsDMARD-experienced patients (c). NRI results are depicted by filled circles, and as-observed results are depicted by hollow circles. Results are statistically significant if 1 is not covered by the 95% CI for the odds ratios. Remission is defined as CDAI score ≤ 2.8, and LDA is defined as CDAI score > 2.8 and ≤ 10. Results of the best performing model can be found in Fig. S7. bDMARD biologic disease-modifying anti-rheumatic drug, CDAI Clinical Disease Activity Index, CI confidence interval, FMA frequentist model averaging, LDA low disease activity, NRI non-responder imputation, OMA other mechanism of action, TNFi tumour necrosis factor inhibitor, tsDMARD targeted synthetic disease-modifying anti-rheumatic drug

Full size image

The proportion of patients achieving LDA and remission was consistent across cohorts for both age populations, with slightly greater proportions reported for patients in the ≥ 65 years group for cohort A and cohort B–TNFi for both measures (Table 4).

Other Disease Activity (CDAI, Pain VAS, and HAQ-DI, CFB)

At 24 months, greatest mean CFB in CDAI was reported by those receiving baricitinib in the overall, naïve, and both age groups (Table 4). Using comparative analyses, our data showed that patients treated with baricitinib demonstrated a significantly greater improvement in CDAI CFB compared to TNFi across overall, naïve, and experienced groups for observed and imputed data, with the exception of the experienced group, which was only significant for observed data, and compared to cohort B overall in the overall and naïve groups (Fig. 4). Similarly, significantly greater improvements in CDAI CFB compared to cohort B overall, TNFi, and OMA were noted in ≥ 65 age group and vs. TNFi in the < 65 age group for both observed and imputed data (Fig. S10).

Fig. 4
figure 4

Comparative adjusted analysis using FMA approach of CDAI, Pain VAS, and HAQ-DI for baricitinib versus cohort B overall, TNFi, and OMA at 24 months for overall (a), b/tsDMARD-naïve (b), and b/tsDMARD-experienced patients (c). *Significant differences observed in HAQ-DI. Overall group: barictinib vs. cohort B overall (LSMean difference − 0.075; 95% CI [− 0.123, − 0.012]), barictinib vs. OMA (LSMean difference − 0.094; 95% CI [− 0.152, − 0.015]); Naïve group: baricitinib vs. TNFi (LSMean difference − 0.089; 95% CI [− 0.182, 0.001]); Experienced group: baricitinib vs. OMA (LSMean difference − 0.081; 95% CI [− 0.16, − 0.001]). mBOCF results are depicted by filled circles, and as-observed results are depicted by hollow triangles. Results are statistically significant if 0 is not covered by the 95% CI. Results of the best performing model can be found in Fig. S9. bDMARD biologic disease-modifying anti-rheumatic drug, CDAI Clinical Disease Activity Index, CFB change from baseline, CI confidence interval, FMA frequentist model averaging, HAQ-DI Health Assessment Questionnaire-Disability Index, LSMean least squares mean, mBOCF modified Baseline Observation Carried Forward, OMA other mechanism of action, SD standard deviation, TNFi tumour necrosis factor inhibitor, tsDMARD targeted synthetic disease-modifying anti-rheumatic drug, VAS visual analogue scale (mm)

Full size image

At 24 months, CFB in pain VAS was reported across overall, b/tsDMARD-naïve, and b/tsDMARD-experienced (Table 4). For the comparative analysis, average pain VAS was statistically reduced in patients treated with baricitinib vs. cohort B for the overall (LSMean difference − 3.2; 95% CI [− 6.1, − 0.03]) and experienced (imputed data) (LSMean difference − 3.9; 95% CI [− 7.1, − 0.7]) groups, and when compared with TNFi in the experienced group (LSMean difference − 6.2; 95% CI [− 10.1, − 0.9]) (Fig. 4). In the age subgroups, mean CFB in pain VAS was reported by patients across both groups, with the greatest mean CFB observed with baricitinib (< 65 [− 21.7 (standard deviation [SD] 30.6)] and ≥ 65 [− 29.6 (SD 30.1)]) (Table 4). Patients ≥ 65 treated with baricitinib demonstrated a significantly greater improvement in mean pain VAS CFB, for the imputed data, compared to cohort B overall (LSMean difference − 5.3; 95% CI [− 9.7, − 1.8]) and OMA (LSMean difference − 6.0; 95% CI [− 12.3, − 0.8]) (Fig. S10). These results for patients ≥ 65 were consistent for the observed data.

Differences in HAQ-DI were low between cohorts and across overall, naïve, experienced, and age subgroups; however, significant differences favouring baricitinib were observed for baricitinib vs. cohort B overall (LSMean difference − 0.075, 95% CI [− 0.123, − 0.012]), baricitinib vs. OMA (LSMean difference − 0.094, 95% CI [− 0.152, − 0.015]), naïve-baricitinib vs. TNFi (LSMean difference − 0.089, 95% CI [− 0.182, − 0.001]), and experienced-baricitinib vs. OMA (LSMean difference − 0.081, 95% CI [− 0.16, − 0.001]) (Fig. 4). Patients ≥ 65 treated with baricitinib demonstrated a significantly greater improvement in HAQ-DI CFB compared to cohort B overall, TNFi, and OMA (Fig. S10).

Discussion

The results of this real-world study, performed on a European subset of patients, demonstrate that a lower proportion of patients discontinued RA treatment at 24 months when treated with baricitinib compared to other tsDMARDs and bDMARDs, despite differences in age, disease duration, and previous bDMARD experience between cohorts at baseline. Most often treatments were discontinued across treatments as a result of other reasons, which included patient or physician decision, non-compliance, cannot afford medication, unknown causes, or COVID-19-related causes (medical and non-medical reasons). Findings from this study are in agreement with previous European findings that highlighted an overall higher drug maintenance and retention for JAK inhibitors over TNFi and OMA bDMARDs [21]. Advances in therapy have improved outcomes for patients with RA, but treatment discontinuation is still common, with loss of efficacy and lack of effect reported as common reasons for treatment discontinuation [22].

In this study, drug survival up to 24 months was significantly greater for patients treated with baricitinib, compared to TNFi, non-TNFi, and tsDMARDs, across age and prior experience subgroups with overall reasons for discontinuation including ineffectiveness, adverse events, and other reasons. A somewhat lower proportion of patients on baricitinib discontinued treatment because of ineffectiveness in comparison to other treatment arms, particularly in naïve patients, which is consistent with previous reports demonstrating that baricitinib was less frequently stopped as a result of loss of effectiveness within the first year of treatment compared to TNFi, reinforcing its role in a treat-to-target approach [13, 23]. This is further supported by comparative analyses which demonstrated that baricitinib is more effective than TNFi and OMA in improving CDAI for overall and naïve groups. These findings replicate the RA-BEAM results vs. TNFi in real-world setting, despite older patients with longer disease durations and patients with long-lasting RA [3]. Effectiveness was similar between TNFi and OMA in other measures, except in patients < 65 years where baricitinib showed a significantly greater proportion achieving LDA, compared with TNFi and OMA. Of note, the higher rate of discontinuation due to ineffectiveness and other reasons in cohort B treatment arms may impact the effectiveness outcome results reported.

Average pain scores significantly reduced over time for patients treated with baricitinib vs. cohort B for the experienced patients, in particular when compared with TNFi; however, there was no difference observed when compared with tsDMARDs. Of note, baricitinib significantly improved HAQ-DI CFB in the overall population compared to TNFi, proving earlier interventions with baricitinib significantly improve disease activity and will have a clear benefit on functionality vs. standard of care. On the other hand, for the most difficult to treat/refractory population, experience to b/tsDMARDs shows significantly better LDA and pain VAS control, in particular for baricitinib compared to TNFi [24, 25].

The present results are in agreement with those reported in randomised controlled trials. In fact, RA-BEAM and RA-BEGIN demonstrated that significantly more patients achieved remission and LDA when treated with baricitinib compared to adalimumab and methotrexate, respectively [8]. Additionally, greater improvements in pain and physical function were observed with baricitinib than with adalimumab and placebo [26]. In RA-BE-REAL, a greater proportion of patients achieved the therapeutic target of CDAI remission or LDA at 24 months with baricitinib treatment compared to other tsDMARDs and bDMARDs. Additionally, we report greater improvement in CDAI disease activity, as well as pain and physical functioning. Furthermore, results presented here echo findings described in other real-world studies; patients treated with baricitinib are reported to have high remission and LDA, regardless of treatment history, from 6 to 12 months of treatment [27,28,29], while other reports have noted similar proportions of patients achieving these outcomes at 12 months, regardless of treatment with baricitinib, TNFi, or OMA [30].

Of note, patients treated with a tsDMARD more frequently had experience with at least two previous b/tsDMARDs at baseline, while those in the TNFi group did not as frequently have experience with previous b/tsDMARDs. This may affect the treatment response; however, the statistical approach used for comparative analyses aimed to take into consideration prior treatment with csDMARDs only, prior treatment with bDMARDs/tsDMARDs (split by TNFi only, other non-TNFi bDMARDs, prior tsDMARDs, all), and concomitant use of csDMARDs at baseline.

Strengths

RA-BE-REAL is one of the first large-scale multinational studies generating real-world evidence on the use of baricitinib, bDMARDs, and other tsDMARDs in clinical practice. As a result of stringent study inclusion criteria, randomised controlled trials do not fully reflect the treatment complexities and heterogeneous patient populations observed in real-world practice. Additionally, this study includes the large sample size of 1073 patients across multiple healthcare settings in Europe. An additional strength of this study is the use of a statistical methodology utilising model averaging, and machine learning approaches that have been well established in other areas. The results were consistent across many different statistical analysis strategies per outcome, for both imputed and as-observed analyses reconfirming previous findings from randomised controlled trials (Fig. S2).

Limitations

As an observational study, limitations include patient selection bias, as the choice of treatment with baricitinib, any bDMARD, or any other tsDMARD was based on the judgement of the treating physician. Also, this observational study was not designed to specifically collect data on adverse events and only included reasons for discontinuation related to safety. The potential effect of unmeasured confounding was evaluated using the E-value, which suggested the robustness of the reported treatment estimates (Fig. S11). As this was an observational study, we did not control the population enrolled. Unlike randomised controlled trials there are no strict exclusion criteria to exclude patients from this study.

Conclusion

RA-BE-REAL expands on data already obtained from randomised controlled trial evidence, confirming the significantly higher proportion of patients on treatment and longer time to discontinuation for baricitinib across subpopulations (age and naïve/previous ts/bDMARDs) compared to other tsDMARDs and bDMARDs probably led by enhanced effectiveness over time. Furthermore, discontinuation of baricitinib due to safety was similar to TNFi and non-TNFi across subpopulations. This analysis described treatment of baricitinib in a heterogeneous real-life population, across multiple countries in Europe, and can provide better information about the use of baricitinib.