Influence of the shared epitope on the effectiveness of biologic antirheumatic agents in European patients with rheumatoid arthritis

Abstract

Background

For rheumatoid arthritis (RA), the HLA-DRB1 ‘shared epitope’ (SE) allele has been proposed as a predictive biomarker for the effectiveness of mainly abatacept and other biologic disease-modifying antirheumatic drugs (bDMARDs).

Objective

To explore the relative impact of SE positivity on the effectiveness of different bDMARDs in European patients with RA.

Methods

Nested cohort study in the Swiss (Swiss Clinical Quality Management) and Danish (DANBIO) rheumatology registries. Patients with RA who initiated bDMARD treatment and had available DNA samples to define SE were included. Four bDMARD groups were defined: abatacept, tumour necrosis factor inhibitors (TNFi), rituximab, tocilizumab. Patients were matched 1:1 using propensity scores and the effectiveness compared between patients being SE positive versus SE negative. For sensitivity, the number of SE alleles was included. Primary end point was treatment retention (crude by SE status, adjusted Cox regression analyses including relevant covariates). Secondary end points were low disease activity (LDA) and remission (1, 2 years).

Results

Of the 5248 treatment courses initially identified, 464 patients in each of the four bDMARD groups were matched (SE negative: 31%, one SE allele: 48%, two alleles: 21%). For all bDMARD groups, crude 1-year treatment retention was unaffected by SE status. Cox regression analyses showed non-significant HRs for SE-positive versus SE-negative patients: abatacept 0.99 (95% CI 0.71 to 1.38), TNFi 1.01 (0.66 to 1.54), rituximab 1.14 (0.71 to 1.84), tocilizumab 1.48 (0.92 to 2.40). LDA and remission rates were unaffected by SE status for all four bDMARDs. Sensitivity analyses showed similar results.

Conclusion

In the current study, the SE was not associated with bDMARD effectiveness in patients with RA of mainly European ancestry.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• For rheumatoid arthritis (RA), the HLA-DRB1 ‘shared epitope’ (SE) allele has been proposed as a predictive biomarker for the effectiveness of abatacept and other biologic disease-modifying antirheumatic drugs (bDMARDs).

• This result is based on studies performed mainly in patients of Asian ancestry.

WHAT THIS STUDY ADDS

• In this large observational European cohort, approximately two-thirds of patients were SE positive.

• SE had no influence on either treatment retention or 2-year effectiveness of abatacept, tumour necrosis factor inhibitors, rituximab or tocilizumab.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The SE was not associated with bDMARD treatment effectiveness in patients of European ancestry.

Background

Observational data from registries provide a unique opportunity to understand the clinical effectiveness of disease-modifying antirheumatic agents (DMARDs) in specific clinical situations, which is becoming increasingly important as we move towards more personalised clinical care. Both rheumatoid factor (RF) and anticitrullinated protein antibodies (ACPAs) are being used as diagnostic tools for rheumatoid arthritis (RA), but may also be used as prognostic factors or as predictors of response to therapy, as these biomarkers have been associated with better clinical response to some biologic (b-)DMARDs.^1–3

Genetic markers have also been proposed as predictors of treatment response. A number of variants of the major histocompatibility complex (MHC) are the strongest known genetic risk factors for RA and are associated with more severe disease. The term ‘shared epitope’ (SE) refers to a specific five-amino acid motif within the DR chain encoded by HLA-DRB1 alleles. This motif is notably linked to the susceptibility of developing ACPA-positive RA. This finding has led to the hypothesis that patients carrying the SE variants bind preferentially citrullinated proteins, resulting in the production of ACPAs.^4–7

In previous cohort studies, only abatacept and rituximab, both mostly cell-mediated bDMARDs, displayed significantly better effectiveness in seropositive patients.1 2 8 9 Japanese researchers have suggested that this effect of seropositivity could be mediated by genetics. Oryoji et al observed that the risk of abatacept discontinuation was up to 10 times higher in SE-negative patients and the likelihood of remission was more than twice as high in SE-positive patients.¹⁰ Furthermore, other Japanese authors suggested that SE-positive patients treated with abatacept display larger improvement in disease activity and joint ultrasound scores at 1 year.^{11 12} However, other researchers did not find higher effectiveness of abatacept compared with tumour necrosis factor inhibitors (TNFi) in SE (and ACPA)-positive patients with RA or found only minor differences.^12–16 Thus, the role of SE as a predictor of treatment outcomes for abatacept and other bDMARDs is currently unclear. These inconsistent results, which are often based on smaller cohorts, warrant replication in a larger representative sample of patients with RA. Furthermore, from a genetic perspective, it is relevant to test whether associations mainly found in Asian populations can be confirmed in study populations of other ancestries. Thus, the aim of this study was to explore the relative impact of the SE on the effectiveness of abatacept and other bDMARDs in patients with RA from large observational cohorts from two European countries.

Methods

Study design and study population

This was a nested case-control analysis in two prospective, longitudinal cohorts of patients with RA treated with bDMARDs in real-life settings. Inclusion criteria were: diagnosis of RA, treatment with either abatacept or a bDMARD with other modes of action, prospective follow-up of effectiveness parameters and available genetic material for SE assessment.

In Switzerland, sampling occurs since 2011 within the Swiss RA registry (Swiss Clinical Quality Management (SCQM)). Serum samples and whole blood samples are collected at various participating centres and stored centrally at −80°C. DNA is extracted using standard methodology from the Swiss National Histocompatibility laboratory and aliquoted for each participant. In Denmark, sampling has been conducted since 2015 across >10 rheumatology departments within the Danish Rheumatologic Biobank infrastructure. Materials are stored as serum, plasma and whole blood samples at −80°C in decentralised facilities.¹⁷ Corresponding clinical data are registered in the nationwide DANBIO registry.^18–20

Data protection

Both in Switzerland and Denmark, biological material is available for research purposes in nationwide biobanks.²¹

The study was conducted in accordance with the International Society for Pharmacoepidemiology Guidelines for Good Epidemiology Practices and European Alliance of Associations for Rheumatology (EULAR) points to consider for the analysis and reporting of comparative effectiveness research and applicable regulatory requirements.²²

Exposure of interest

The exposure of interest was the interaction between the SE and bDMARDs. The bDMARDs investigated included four groups with different modes of action: abatacept, TNFi (adalimumab, certolizumab pegol, etanercept, golimumab, infliximab), anti-CD20 (rituximab), anti-interleukin-6 (anti-IL6) (tocilizumab). Individual patients could contribute multiple bDMARD treatment courses.

The presence of one or two HLA-DRB1 alleles was defined as SE positivity. SE determination was performed separately in each country (Switzerland, Denmark) to comply with data protection regulations. In Switzerland, HLA-DRB1 polymorphisms were assessed by reverse PCR sequence-specific oligonucleotide primers (SSP) hybridisation and by PCR sequence-specific primers using commercial reagents validated by the respective laboratories (at the National Reference Laboratory for Histocompatibility in Switzerland). The Score software tool was used to perform a standardised evaluation of the HLA typing results with the Olerup SSP kits. Within the HLA-DRB1*04 haplotype, the method discriminates all major subtypes in different allele groups. HLA-DR1, DR4 and DR14 alleles that are negative for the SE70-74 motif could be discriminated. In a second step, the SE-positive typing ambiguities were analysed by PCR-SSP in order to determine the final four-digit typing result. For the Danish samples, genotyping was performed at deCODE genetics on the Illumina platform with the Global Screening Assay²³ and HLA-DRB1 SE positivity was defined using the Graphtyper software,²⁴ pooling the following HLA-DRB1 alleles: 01:01, 01:02, 04:01, 04:04, 04:05, 04:08, 10:01, 14:02, as previously described.²⁵ We assessed the relative effectiveness of the medications in patients according to SE status (positive, negative).

Outcomes

The primary end point was treatment retention rate, defined as the time from treatment initiation (baseline) to treatment discontinuation. In cases where treatment discontinuation was ambiguous (ie, rituximab), a previously published heuristic was employed to determine the treatment stop date: treatment was considered discontinued at either the registered stop date, the date of initiation of a new bDMARD, the date of initiation of a new triple therapy of conventional synthetic (cs-)DMARDs or the date of death, whichever occurred first.²⁶

Secondary outcomes included the rates of low disease activity (LDA) and remission, at 1 year and 2 years, using the Clinical Disease Activity Index (CDAI). CDAI was used instead of DAS28 to avoid an assessment bias in favour of medications, which have a strong impact on acute phase reactants (ie, anti-IL6 agents).²⁷ The following time window was used to identify measures: ±3 months.

Data censoring occurred 1 December 2022.

Confounding and covariates

To minimise the risk of confounding, a two-step process was applied. First, patients treated with abatacept were matched 1:1:1:1 to patients who initiated TNFi agents, a third group who initiated rituximab and a fourth group who initiated tocilizumab. We used a propensity score to perform the matching, with the aim to select patients who were overall similar across countries. Second, differences in covariates were addressed by multivariable regression analyses (see ‘Data analyses’ section).

A patient was categorised seropositive if one or both RF and ACPA were positive. A patient was categorised seronegative if both RF and ACPA were negative, or one biomarker was missing and the other was negative.

Further baseline clinical variables were identified in SCQM and DANBIO and included: age, sex, calendar year, previous number of b/targeted synthetic (ts-) DMARD treatments (0/1/2/≥3), disease duration, concomitant use of csDMARDs (yes/no, grouped as: methotrexate, methotrexate and other csDMARD, none, other (leflunomide, sulfasalazine, hydroxychloroquine)), concomitant glucocorticoid (yes/no), disease activity (CDAI, Disease Activity Score 28 (DAS28)), comorbidities (yes/no for any of lung disease, cardiovascular disease, hyperlipidaemia, diabetes, neuropsychiatric disorder), smoking (current/previous/never smoker).

Data analyses

A prespecified evaluation of effect modification across countries was applied; data pooling was performed only if no significant heterogeneity was observed regarding the impact of SE on relative drug effectiveness across registries. A propensity score was generated using multiple logistic regression based on the following baseline variables: calendar year, number of previous b/tsDMARDs, sex and country. These variables were selected based on their a priori relevance and on their high completeness (100% availability) to prevent patient loss at this stage. We then performed 1:1 nearest-neighbour matching using a calliper of 0.2 to achieve satisfactory balance.

Standard descriptive statistics were used to summarise baseline characteristics stratified by bDMARD group.

We investigated the potential for effect modification by the SE on treatment retention across countries, using Cox proportional hazards models. Since country is a strong driver of drug retention,²⁸ we first examined if there was significant effect modification of the association between SE and relative drug effectiveness by country. Two models were run: the first included both the SE (positive/negative) and country (Switzerland/Denmark) as variables, whereas the second model also included an interaction term between SE and country. We then assessed the significance of the added interaction terms to compare the two models.

Unadjusted treatment retention rates were plotted using the Kaplan-Meier method and differences were assessed using the log-rank test. Adjusted treatment retention rates were analysed using Cox proportional hazard models for matched datasets,²⁹ adjusting for potential confounding factors (see ‘Confounding and covariates’ section), including an interaction term between bDMARD group and SE. Matching variables were also included as covariates to account for residual confounding, except sex, which was too collinear with treatment. A cluster term by patient was introduced to account for the fact that a patient could have received several bDMARDs at different time periods.

Missing baseline covariates were imputed using multiple imputations with chained equations (50 samples and 25 iterations, predictive mean matching algorithm for continuous variables and logistic and polytomous regression for categorical variables). All baseline covariates were used in the imputation. Treatment effectiveness was assessed through CDAI remission (≤2.8) or LDA (≤10) at 1 and 2 years. The confounder-adjusted response rate with attrition correction³⁰ was used to obtain adjusted and unbiased estimates of differences in LDA and remission rates between SE-positive and SE-negative patients. This methodology is specifically designed to address the challenges of attrition, by including treatment cessation reasons in the multiple imputation method.

All statistical analyses were performed with R V.4.3.1 (R Core Team, 2023, Vienna, Austria).

Sensitivity analyses

For treatment retention, analysis including only ACPA-positive patients was performed. Furthermore, we considered the number of SE alleles (none, one, two) to investigate a potential dose effect on effectiveness. Furthermore, analyses including all treatment courses (no matching) were performed.

Results

A total of 5248 treatment courses with SE information were obtained for this study. Data from Switzerland and Denmark were pooled for analyses, as no effect modification by country was observed in analyses of the association between SE and relative drug effectiveness (details not shown). Following propensity score matching, each of the four treatment groups was balanced with 464 courses (figure 1). All standardised mean differences for baseline characteristics were below 0.22 after matching (online supplemental table S1). SCQM contributed approximately two-thirds of the treatment courses, and DANBIO the remaining one-third. The proportion of SE-positive patients was 67.1% in Switzerland and 70.6% in Denmark, respectively, with two SE alleles being more frequent in Denmark (online supplemental table S2). The number of SE alleles was similar between the four bDMARD groups (table 1). Patients on rituximab tended to have longer disease duration and were more often seropositive. Baseline disease activity was higher in patients treated with abatacept and tocilizumab. Patients with one or two SE alleles had longer disease duration and were, as expected, more often seropositive. SE-negative patients presented higher baseline disease activity (CDAI, DAS28) but lower CRP (online supplemental table S2). For baseline characteristics, missingness was low for serostatus, disease duration and CRP, but high for baseline CDAI and DAS28 (table 1), irrespective of treatment group and number of SE alleles (online supplemental table S3). Reason for treatment discontinuation was mainly ineffectiveness or other reasons and was numerically similar for SE-positive and SE-negative patients (online supplemental table S4). The crude treatment retention rates in the four bDMARD groups, stratified by SE status, are displayed in figure 2. We found no statistically significant differences between SE-positive and SE-negative patients across all four groups (log rank >0.05). In sensitivity analyses, results were similar in the subgroup of patients who were ACPA positive (online supplemental figure S1), when considering the number of SE alleles (online supplemental figure S2), and when taking into account all treatment courses (n=5248 before matching, of which n=4708 belonged to one of the four bDMARD groups) (online supplemental figure S3).

Figure 1.

Study design: propensity score matching. Flow chart of the propensity score matching. 1:1 nearest-neighbour matching was performed on the following baseline variables: calendar year, line of therapy (previous biologic or targeted synthetic disease-modifying antirheumatic drug treatments), sex and country. ABA, abatacept; RTX, rituximab; TCZ, tocilizumab; TNFi, tumour necrosis factor inhibitors.

Table 1.

Baseline characteristics stratified by bDMARD group, pooled for Switzerland and Denmark

	Abatacept (n=464)	Rituximab (n=464)	Tocilizumab (n=464)	TNFi (n=464)	Missing (%)
Country (Denmark)	155 (33.4)	148 (31.9)	166 (35.8)	160 (34.5)	0.0
Treatment duration (median (IQR))	1.8 (0.6, 5.9)	3.6 (1.0, 7.5)	2.3 (0.6, 7.0)	1.0 (0.4, 2.8)	0.0
Age, years	56.9 (12.7)	57.7 (12.3)	56.7 (13.3)	55.4 (13.3)	0.0
Sex (female) (%)	357 (76.9)	348 (75.0)	368 (79.3)	357 (76.9)	0.0
Disease duration, years (median (IQR))	9.5 (4.7, 17.4)	11.1 (5.6, 18.6)	10.0 (5.2, 18.0)	9.6 (4.4, 16.9)	1.3
Seropositivity (ACPA/RF) (%)	335 (73.1)	369 (81.1)	324 (71.4)	296 (66.1)	2.2
ACPA positive (%)	300 (65.8)	314 (71.0)	284 (63.7)	252 (56.6)	3.4
RF positive (%)	294 (65.9)	299 (67.3)	295 (65.8)	266 (59.8)	3.7
Shared epitope, number of alleles (%)					0.0
0	132 (28.4)	129 (27.8)	157 (33.8)	154 (33.2)
1	229 (49.4)	245 (52.8)	226 (48.7)	200 (43.1)
2	103 (22.2)	90 (19.4)	81 (17.5)	110 (23.7)
Previous b/tsDMARDs (%)					0.0
0	91 (19.6)	64 (13.8)	80 (17.2)	108 (23.3)
1	104 (22.4)	127 (27.4)	113 (24.4)	107 (23.1)
2	123 (26.5)	127 (27.4)	113 (24.4)	116 (25.0)
3 or more	146 (31.5)	146 (31.5)	158 (34.1)	133 (28.7)
Concomitant csDMARDs (%)					0.0
MTX	200 (43.1)	184 (39.7)	133 (28.7)	196 (42.2)
MTX+other	23 (5.0)	20 (4.3)	23 (5.0)	27 (5.8)
None	156 (33.6)	168 (36.2)	239 (51.5)	149 (32.1)
Other	85 (18.3)	92 (19.8)	69 (14.9)	92 (19.8)
Concomitant GC (%)	155 (33.5)	162 (35.3)	137 (29.7)	134 (29.2)	0.8
CRP (mg/L)	11.5 (16.3)	11.5 (19.1)	10.7 (22.2)	11.2 (19.4)	23.7
CDAI	23.1 (12.2)	20.8 (11.1)	23.1 (12.2)	19.9 (11.8)	68.9
DAS28	4.5 (1.3)	4.4 (1.2)	4.6 (1.2)	4.2 (1.3)	64.8
Smoking (%)					66.7
Current	29 (19.0)	26 (18.2)	25 (15.2)	22 (14.0)
Never	63 (41.2)	62 (43.4)	91 (55.2)	70 (44.6)
Past	61 (39.9)	55 (38.5)	49 (29.7)	65 (41.4)
Comorbidity (%)	112 (24.1)	123 (26.5)	101 (21.8)	86 (18.5)	0.0

Numbers are mean (SD) unless otherwise stated.

ACPA, anticitrullinated protein antibody; bDMARDs, biologic DMARDs; CDAI, Clinical Disease Activity Index; CRP, C reactive protein; csDMARDs, conventional synthetic DMARDs; DAS28, Disease Activity Score 28; DMARDs, disease-modifying antirheumatic drugs; GC, glucocorticoids; MTX, methotrexate; RF, rheumatoid factor; TNFi, tumour necrosis factor inhibitors; tsDMARDs, targeted synthetic DMARDs.

Figure 2.

Crude treatment retention according to shared epitope (SE) status (positive/negative) and biologic disease-modifying antirheumatic drug (bDMARD) group (n=464 per bDMARD group). SE positive in blue, SE negative in red. Crude retention rate estimated by Kaplan-Meier curves, stratified by SE positivity. P values from log-rank test. X-axis shows number of years. TNF-inhibitors, tumour necrosis factor inhibitors.

The association of the SE with treatment retention did not differ significantly across bDMARD groups (test for interaction, adjusted p value >0.05). The adjusted HRs of discontinuing the bDMARD for SE-positive patients compared with SE-negative patients were 0.99 (95% CI 0.71 to 1.38) for abatacept, 1.01 (95% CI 0.66 to 1.54) for TNFi, 1.14 (95% CI 0.71 to 1.84) for rituximab and 1.48 (95% CI 0.92 to 2.40) for tocilizumab (figure 3, online supplemental table S5). In sensitivity analyses, we found no significant association with treatment retention when we considered number of SE alleles (online supplemental table S6).

Figure 3.

Treatment discontinuation in SE-positive versus SE-negative patients stratified by biologic disease-modifying antirheumatic drug (bDMARD) group. Results from multivariable Cox regression analyses. Analyses were adjusted for the following baseline factors: age, history of b/tsDMARD treatment(s), disease duration, seropositivity, concomitant therapy with conventional synthetic DMARDs and glucocorticoids, disease activity, any comorbidity and calendar year with a strata term for country and a cluster term for patient to account for patients contributing multiple bDMARD courses. ABA, abatacept; TNFi, tumour necrosis factor inhibitors; RTX, rituximab; SE, shared epitope; TCZ, tocilizumab.

Secondary effectiveness outcomes were the rates of LDA and remission at 1 and 2 years. The rates of remission and LDA at 1 year were not significantly different in SE-positive patients compared with SE-negative patients in any of the four bDMARD groups (table 2).

Table 2.

Differences in adjusted response rates between SE-positive compared with SE-negative patients, by bDMARD groups

		Abatacept	Rituximab	Tocilizumab	TNFi
1 year	LDA (CDAI ≤10)	+2.9% (−17.4% to 23.2%)	−10.1% (−25.8% to 5.6%)	−10.1% (−25.8% to 5.6%)	+10.3% (−9.2% to 29.9%)
	Remission (CDAI ≤2.8)	+2.8% (−6.2% to 12.0%)	−2.8% (−14.2% to 8.6%)	−1.0% (−14.7% to 12.6%)	+11.3% (−7.6% to 30.2%)
2 years	LDA (CDAI ≤10)	+4.6% (−16.0% to 25.3%)	−16.5% (−32.4% to 0.6%)	−11.2% (−28.4% to 6.1%)	−0.3% (−17.7% to 17.2%)
	Remission (CDAI ≤2.8)	+4.7% (−9.0% to 18.5%)	−10.5% (−28.4% to 7.4%)	−3.8% (−17.1% to 9.4%)	−1.2% (−15.0% to 12.7%)

Adjusted CDAI LDA and remission rates differences for SE-positive compared with SE-negative patients, with 95% CIs, by bDMARD group.

Analyses were adjusted for the following baseline characteristics: concomitant therapy with conventional synthetic DMARDs, concomitant glucocorticoid, calendar year, number of previous b/tsDMARD treatment(s), treatment duration, disease activity, disease duration, age and sex, as well as for attrition.

b/tsDMARD, biologic and/or targeted synthetic disease-modifying antirheumatic drug; CDAI, Clinical Disease Activity Index; LDA, low disease activity; SE, shared epitope; TNFi, tumour necrosis factor inhibitors.

In a post hoc analysis, we explored a potential association with the HLA-DRB1*04:05 allele in the Swiss subcohort, but did not find a significant association with bDMARD effectiveness (data not shown).

Discussion

With this large-scale observational study based on collaborative results from Switzerland and Denmark, we aimed to explore if previous studies, primarily conducted in Asian RA populations, which had shown beneficial results of SE on abatacept treatment outcomes, could be replicated in a European population. We did not demonstrate any significant impact of SE status on treatment retention or proportion of responders for neither abatacept nor other bDMARD groups.

Several theories have been suggested to explain the association between SE and RA. Since MHC class II molecules play a crucial role in presenting peptides to T helper (Th) lymphocytes, it is theorised that individuals with SE may preferentially present arthritogenic peptides, in particular citrullinated peptides. The SE epitope amino acid motif encodes one of the peptide-binding grooves and appears to have a stronger affinity for citrullinated peptides.³¹ Other theories have been proposed, including the possibility of a pro-oxidative effect triggered by nitric oxide,³² and an increased polarisation of lymphocytes towards Th17 cells.³³

Given that the presence of the SE is closely linked to the presence of ACPAs, and ACPAs are known to be associated with a better response to certain bDMARDs like rituximab and abatacept,^{1 34} one might anticipate that SE-positive patients would exhibit better responsiveness to these agents. So far, this pattern has mainly been confirmed in smaller Japanese studies.^{10 12} Japanese results reporting a lower risk of abatacept discontinuation and a higher likelihood of achieving remission in SE-positive patients were explored in a subsequent study (106 patients treated with a first-line bDMARD, of whom 37 were treated with abatacept). In that study, mainly the HLA-DRB1*04:05 allele within the SE improved effectiveness to abatacept treatment.³⁵ Such an association was not confirmed in a post hoc analysis of our current study.

Similar to our study findings, recent studies performed outside Japan have not been able to demonstrate convincing beneficial effects of SE for bDMARD effectiveness in RA.^{12–16 36} The reasons for this inability to replicate results in patients with RA of non-Japanese ancestry remain somewhat uncertain. One potential explanation might be differences in the distinct genetic backgrounds, which may lead to differing impact of the SE across ethnic groups. In our study, we found very consistent outcomes by SE status in both the Swiss and the Danish patient cohorts (data not shown), which suggest that these findings are robust and not influenced by specific national factors. Furthermore, it is worth highlighting that the present analysis was conducted with a considerably larger sample size, significantly reducing the risk of spurious associations. Other contributing factors might be differences in study design and definition of study outcomes. Thus, in the study by Harrold et al, the impact of SE and ACPA positivity depended on matching procedures and prior exposure to bDMARDs. Of interest, a recent Korean study indicated relevance of HLA-DRB1 locus changes outside the SE region.⁹

ACPA status is closely associated with SE. Of interest, our results were robust in a subanalysis including only ACPA-positive patients. It was beyond the scope of the study to explicitly explore the impact of seropositivity on treatment outcomes as this has been explored previously.³⁷ The well-known interplay between smoking and SE in ACPA-positive patients has been confirmed previously.³⁸ Thus, for the current study, smoking was included as a covariate in confounder-adjusted analyses to identify SE effects independent of smoking status.

This study has some limitations that need to be acknowledged. To increase statistical power, data were pooled across two countries. This approach was supported by analyses showing no interaction/effect modification³⁹ on the effect of the SE on the relative treatment effectiveness by country. However, country-specific effects and differences in the occurrence of SE alleles could have affected results, although testing of the latter confirmed similar findings. As a post hoc analysis, we explored the country-specific main effect of the SE on treatment retention and confirmed similar findings, namely HR 0.91 (95% CI 0.67 to 1.24) and 1.12 (0.66 to 1.89) for Switzerland and Denmark, respectively. Although it might have been relevant to explore treatment group estimates per country, we refrained from doing so in order to avoid low power and spurious findings.

Missing data and incomplete follow-up was a challenge, as usual for most registries.²² Thus, two-thirds patients had missingness for composite measures of baseline disease activity, with slightly less missing data (between 5% and 10%) for tocilizumab (data not shown). Similarly, missingness was high for LDA and remission. This was addressed by robust statistical methods to compare response rates, taking into account, for example, attrition bias. Reassuringly, missingness appeared similar across treatment groups and number of SE alleles. However, for precaution, measures of disease activity were applied only as secondary and not primary outcomes to address this limitation. Drug retention is considered a robust outcome for observational studies, and here data were complete. Stratification based on reason for discontinuation (eg, lack of efficacy vs adverse events) was not performed due to withdrawal reasons often being stated as ‘other’ (>40%). This illustrates the complexity of routine care treatment where withdrawals often occur due to several contributing reasons. Furthermore, we did not have any specific hypothesis regarding the mechanism behind any potential SE-mediated effectiveness. Therefore, overall withdrawal was considered an acceptable measure.

As with all non-randomised studies, a potential for confounding by indication exists if the choice of a particular drug or predictor of interest is linked to the outcome. However, in this case, the SE status was not available to clinicians when initiating bDMARD therapy and therefore could not have influenced results. To accommodate confounding, we applied a two-step approach. First, in a nested case-control design, we used a matching procedure based on key descriptive variables (sex, country, calendar year) traditionally used in epidemiology research⁴⁰ further adding line of treatment, a known confounding factor. All matching variables had no missing data, thus ensuring minimal patient loss and avoiding imprecise matching. Second, we performed multiple imputation and regression analyses to adjust for most known confounders.⁴⁰ All variables were selected based on a priori reasoning. Missingness was high for baseline CDAI and DAS28. This was accommodated by imputation, where previous studies have shown robustness under the missing-at-random assumption.³⁰ However, it is important to highlight that residual imbalances despite matching and adjustment and confounding cannot be ruled out within our observational setting. For instance, differential missingness across treatments and by SE status might have affected our results. The strength of this work lies in the wide variety of patients typically included in observational studies with heterogeneous disease presentation representing both primary and secondary care and representative of the ‘real world’ and the statistical power stemming from the large number of patients.

In conclusion, we did not find the SE to be associated with bDMARD effectiveness in patients with RA of mainly European ancestry.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information. NA.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

Biological materials were collected after written informed consent and in accordance with ethical and regulatory approvals (Switzerland: GE 10-089; Denmark: The Biomarker Protocol H-2-2014-086, supplementary protocol 49419, Danish Data Protection Agency RH-2015-297). In Switzerland, project-specific ethical approval was obtained before the genetic assessments were conducted and the data analysed. In Denmark, ethical approval was obtained through The Biomarker Protocol and project-specific data processor agreements (Capital Region, P2020-104). All participants gave informed consent before taking part in the study.