The impact of autoantibodies on the efficacy of biological disease-modifying anti-rheumatic drugs in rheumatoid arthritis: meta-analysis of randomized controlled trials

Abstract

Objective

To investigate the efficacy of bDMARDs in patients with RA with RF/ACPA compared with patients without these autoantibodies.

Methods

Previous systematic literature reviews performed by EULAR RA management task forces were searched for qualifying RCTs. RCTs investigating the efficacy of bDMARDs and including both autoantibody-positive (≤80% of total population) and -negative RA patients were eligible. For trials comparing bDMARD+csDMARD vs csDMARD, relative risks (RR) comparing two groups (RF+ vs RF-, ACPA+ vs ACPA-) were calculated for efficacy outcomes for each arm. Subsequently, relative risk ratios (RRRs) were computed, as the ratio of RR of the bDMARD-arm and the RR from the non-bDMARD-arm. Pooled effects were obtained with random effect meta-analyses.

Results

Data from 28 eligible RCTs were analysed, pooling 23 studies in three subgroups: six including csDMARD-naive patients, 14 csDMARD-IR and three TNFi-IR patients. In csDMARD-naive and csDMARD-IR patients, seropositivity was not associated with a better response to bDMARDs: pooled 6-month ACR20 RRRs 1.02 (0.88–1.18) and 1.09 (0.90–1.32), respectively. Other outcomes showed no difference between groups either. In TNFi-IR patients, based on three trials, the 6-month ACR20 RRR was 2.28 (1.31–3.95), favoring efficacy in seropositive patients. Other outcomes mostly showed no significant difference between the groups. Based on the mode of action, efficacy was comparable between RF-positive and RF-negative patients for both TNFi and non-TNFi treatment and also for the individual bDMARDs.

Conclusion

The effect of bDMARDs is generally comparable in patients with and without RF/ACPA, regardless of the patient population, the mechanism of action or individual drug used.

rheumatoid arthritis, biological therapies, rheumatoid factor, anti-citrullinated protein autoantibodies

Rheumatology key messages

• We performed a meta-analysis based on a systematic literature review including solely data from randomized controlled trials to investigate whether the efficacy of bDMARDs differs in autoantibody-positive RA patients compared to those without autoantibodies.

• In both csDMARD-naive and csDMARD-IR patients, the effect of bDMARDs is generally comparable in patients with and without autoantibodies. In TNFi-IR patients, bDMARDs may be more effective in the autoantibody-positive patients, but the low number of trials precludes a firm conclusion.

• Altogether, the effect of bDMARDs is generally comparable in patients with and without RF/ACPA, regardless of the patient population, mechanism of action or individual drug used.

Introduction

Rheumatoid factor (RF) and anticitrullinated protein antibodies (ACPA) are clinically valuable biomarkers for the diagnosis and prognosis of rheumatoid arthritis (RA). Seropositive RA are known to have worse long-term clinical and radiographic outcomes than seronegative RA [1–3], which is an entity with a higher risk of misdiagnosis. Furthermore, based on differences in associations with genetic and environmental risk factors, seropositive and seronegative disease are presumed to have different underlying pathophysiological mechanisms [4]. However, whether serological status may also affect treatment responses to biological disease-modifying anti-rheumatic drugs (bDMARDs) is not completely clear.

Observational data from several biological drug registries have reported that seropositive patients have better treatment responses to rituximab (RTX) and abatacept than seronegative patients [5–12]. This could potentially be due to pathophysiological differences, with B cells playing a larger role in seropositive disease, and thus RTX as anti-B-cell agent, and abatacept, as blocker of costimulation required for optimal activation of T cells and enabling T-cell-help to B cells, being more effective in this disease subset. A meta-analysis of randomized controlled trials (RCTs) with RTX also found that seropositive patients respond better than seronegative patients [13]. However, the only outcomes reported were the change from baseline in the 28-joints Disease Activity Score (DAS28) and ACR50, raising the question of whether the higher efficacy of RTX observed in the seropositive subset was consistent across different outcome measures. Moreover, the efficacy of RTX in seropositive vs seronegative patients was reported separately for the RTX and placebo groups, without a formal comparison between them. Regarding clinical response to tumor necrosis factor inhibitor (TNFi), a meta-analysis has been conducted on its association with serological status, including 14 studies. In this report, the status of RF or ACPA was not associated with the clinical response to TNFi [14]. However, this analysis was mainly based on data from observational studies, in which a risk of bias in treatment allocation cannot be excluded.

There is therefore still uncertainty regarding the efficacy of bDMARDs in seropositive compared with seronegative patients. In the era of personalized medicine, it would be very helpful to identify factors (for example, the presence of autoantibodies) that can assist the choice of the best suitable treatment for the patient. Our aim was therefore to investigate whether the efficacy of bDMARDs differed between seropositive and seronegative patients. By means of a systematic literature review (SLR) with a meta-analysis of RCTs, we aimed to avoid the potential biases of observational studies and obtain a valid estimate of this comparative effect, enabling us to draw more reliable and solid conclusions.

Methods

Literature search and trial selection

To identify the eligible studies, we made use of previous SLRs addressing the efficacy of bDMARDs performed to inform the task force responsible for the EULAR recommendations for the management of RA published in 2010, updated in 2013 and 2016 [15–17]. The search of the included SLRs was performed from 1962 to February 2016 using MEDLINE, EMBASE and Cochrane CENTRAL databases. From these, the eligible studies were selected by KT-M with the help of S.R. and D.vdW. The following inclusion criteria were used: (i) patients with RA (fulfilling the 1987 ACR [18] or 2010 ACR/EULAR classification criteria [19]) or undifferentiated arthritis; (ii) intervention with one of eight bDMARDs: infliximab (3 mg/kg every 8 weeks), etanercept (50 mg once weekly), adalimumab (40 mg every 2 weeks), certolizumab-pegol (200 mg every 2 weeks), golimumab (50 mg every four weeks), abatacept (∼10 mg/kg according to weight range), RTX (two doses of 1000 mg biweekly) and tocilizumab (8 mg/kg every four weeks) in the on-label dose; (iii) any comparator with or without placebo; (iv) including outcomes on disease activity, functional impairment and/or radiographic progression; (v) phase III RCTs; (vi) trial duration ≥6 months; (vii) ≥50 patients; (viii) trials including both seropositive (RF or ACPA positive) as well as seronegative patients, with a percentage of seropositive patients ≤80% in the intention-to-treat population; and (ix) publications in English. Strategy trials were not included as, due to their complex design, they do not allow one to properly investigate the effect of bDMARDs in subgroups of patients (seropositive vs seronegative) without the confounding effect of the steered-treatment [20]. Trials evaluating IL-1 blockers were excluded because these drugs are currently rarely being used for RA. When the autoantibody status (proportion of patients who were autoantibody-positive vs negative) was not mentioned in the potentially eligible studies, we contacted the authors and/or trial sponsors to assess the eligibility of the trial for our analysis.

Access to additional study data and assessment of risk of bias

As efficacy data stratified for autoantibody status has not been published in the literature systematically, we actively contacted authors of the eligible trials to obtain those data. For each trial fulfilling the inclusion criteria, a group-level data request regarding measures of disease activity, function and radiographic progression stratified by autoantibody status was sent to the first and corresponding authors of the trial, and/or to the trial sponsors. For several of the trials we were directed to international repositories of trial data, (ClinicalStudyDataRequest.com (CSDR) [21], Yale Open Data Access (YODA) [22] and Vivli [23]), through which we obtained access to the individual trial data. A standardized data extraction sheet was provided for the collection of the requested data. ACR20 was the primary end point of this SLR as it was also most frequently the primary end point of RCTs. Secondary endpoints were ACR50, ACR70, DAS28-ESR remission (<2.6) [24], DAS28-CRP remission (<2.6) [24], Health Assessment Questionnaire Disability Index (HAQ-DI) [25] response (≥0.22 improvement), delta DAS28-ESR, delta DAS28-CRP, delta HAQ-DI, and delta total Sharp van der Heijde score (SvdH) [26]. We collected data from trials with the primary end point up to 24 weeks. These outcomes were stratified by baseline autoantibody status (RF+ vs RF- and ACPA+ vs ACPA-). We also aimed at collecting outcomes stratified by both autoantibodies, but these data were only infrequently available and are therefore not analysed. In addition, baseline characteristics (age, gender, DAS28-ESR, DAS28-CRP, HAQ-DI and SvdH) were collected from each trial stratified by serological status and treatment arm. All analyses were performed in the intention-to-treat population. In some of the trials, missing values were imputed, following the procedures of the main analysis of the trial. For dichotomous outcomes, missing data were imputed using non-responder imputation, which assigns a subject with a missing binary or categorical outcome as if they are a non-responder. For continuous variables, missing values were analysed using multiple imputation. Some authors preferred to share individual patient-level data rather than analysed, grouped data. In that case, the first step was to calculate aggregated means and proportions according to the autoantibody pre-defined groups, as described above. The risk of bias of each included trial was assessed using The Cochrane Collaboration’s tool for RCTs [27].

Statistical analysis

Trials were grouped according to the interventions assessed: (I) RCTs of bDMARDs+ conventional synthetic (cs)DMARDs vs csDMARDs, with or without placebo, stratified into the following patient populations, (i) no prior methotrexate use (MTX-naive); (ii) inadequate response to MTX (MTX-IR) and inadequate response to csDMARD (csDMARD-IR); and (iii) TNFi-IR; (II) RCTs evaluating bDMARD monotherapy, either comparing bDMARD in combination with MTX vs bDMARD monotherapy, or comparing bDMARD monotherapy to MTX monotherapy; and (III) head-to-head RCTs of bDMARDs.

Per trial, bDMARD-arm and non-bDMARD-arm (treatment arms) were compared. Relative risks (RRs) and corresponding standard error (SEs) were calculated per treatment arm for dichotomous outcomes. RRs reflect a ratio of the outcomes achieved in the seropositive and seronegative groups. Subsequently, relative risk ratios (RRRs) were computed, as the ratio of the RR of the bDMARD-arm and the RR from the non-bDMARD-treated arm, reflecting whether seropositivity (vs seronegativity) preferentially affected treatment response to bDMARD therapy vs the comparator; that is, whether treatment interacts with autoantibody status. For continuous outcomes, mean differences and their Ses were computed per treatment arm. Subsequently, differences of differences (DoDs) were computed, with values differing from zero again indicating an interaction.

To be able to conduct a meta-analysis, Ses of RRRs and DoDs were computed using the δ-method [28]. Finally, a random-effects meta-analysis was conducted in subgroups of patients defined above and allowing a comparison of the efficacy of bDMARDs according to baseline autoantibody positivity. In order to evaluate the impact of the mode of action (TNFi vs non-TNFi) and individual bDMARDs, a set of sensitivity analyses was conducted in the corresponding subgroups and in all patients, regardless of the previous (b)DMARD exposure. In all meta-regression models, statistical heterogeneity across studies was evaluated with the I² statistic. Data were analysed using R statistical software (version 4.1.1; https://www.r-project.org/).

Results

In total, 321 references of RCTs included in the EULAR RA recommendations bDMARD SLRs [15–17] were screened, yielding 28 RCTs meeting inclusion criteria (see Table 1, Supplementary Fig. S1 and Supplementary References 1–43, available at Rheumatology online). The final analysis set was composed of 23 RCTs of bDMARDs with csDMARDs vs csDMARDS (±placebo), three RCTs of bDMARD monotherapy and two head-to-head RCTs of bDMARDs. Table 1 provides an overview of the included trials. Risk of bias was considered ‘low’ for nearly all trials. Trial details and extracted efficacy outcome measures are listed in the online supplementary section (Supplementary Tables S1.1–S10.4, available at Rheumatology online).

Table 1.

Summary of included trials

Author	Trial name	Patient population	Control/Intervention	ACPA-status known (n)	ACPA-positive (n, %)	RF-status known (n)	RF-positive (n, %)	Risk of bias
A. RCTs of bDMARDs+csDMARDs vs csDMARDs
1. MTX-naive or csDMARD-naive
Emery 2008	COMET	MTX-naive	MTX	172	117 (68%)	NR	NR	Low
			ETN + MTX	205	136 (66%)	NR	NR
Nam 2014	EMPIRE	csDMARD-naive	MTX	52	42 (81%)	54	30 (56%)	Low
			ETN + MTX	54	39 (72%)	55	31 (56%)
Detert 2013	HIT-HARD	csDMARD-naive	MTX	84	44 (52%)	84	58 (69%)	Low
			ADA + MTX	84	46 (55%)	84	57 (68%)
Nam 2014	IDEA	csDMARD-naive	MTX	52	39 (75%)	56	34 (61%)	Low
			IFX + MTX	50	32 (64%)	55	27 (49%)
Leirisalo-Repo 2013	NEORACo	csDMARD-naive	DMARDs	45	34 (76%)	45	33 (73%)	Low
			IFX + DMARDs	46	35 (76%)	46	36 (78%)
Vollenhoven 2009	Swefot	csDMARD-naive	DMARDs	126	71 (56%)	129	84 (65%)	Higha
			IFX + MTX	116	80 (69%)	127	88 (69%)
2. MTX-IR or csDMARD-IR
Kremer 2006	AIM	MTX-IR	MTX	NR	NR	198	171 (86%)	Low
			ABT + MTX	NR	NR	397	349 (88%)
Choy 2012		MTX-IR	MTX	NR	NR	114	93 (82%)	Low
			CZP + MTX	NR	NR	119	89 (75%)
Kay 2008		MTX-IR	MTX	NR	NR	34	27 (79%)	Low
			GLM + MTX	NR	NR	32	28 (88%)
Kim 2007		MTX-IR	MTX	NR	NR	63	52 (83%)	Low
			ADA + MTX	NR	NR	65	50 (77%)
Smolen 2008	OPTION	MTX-IR	MTX	NR	NR	204	144 (71%)	Low
			TCZ + MTX	NR	NR	205	171 (83%)
Keystone 2008	RAPID1	MTX-IR	MTX	NR	NR	198	164 (83%)	Low
			CZP + MTX	NR	NR	392	312 (80%)
Smolen 2009	RAPID2	MTX-IR	MTX	NR	NR	124	97 (78%)	Low
			CZP + MTX	NR	NR	240	186 (78%)
Emery 2010	SERENE	MTX-IR	MTX	168	137 (82%)	172	129 (75%)	Low
			RTX + MTX	167	138 (83%)	170	125 (74%)
Behrens 2021	AMARA	csDMARD-IR	LEF	47	28 (60%)	47	25 (53%)	Low
			RTX + LEF	92	51 (55%)	92	54 (59%)
Smolen 2015	CERTAIN	csDMARD-IR	DMARD	97	56 (58%)	96	66 (69%)	Low
			CZP + DMARD	94	62 (66%)	94	71 (76%)
Combe 2006		csDMARD-IR	SSZ	NR	NR	48	33 (69%)	Low
			ETN + SSZ	NR	NR	96	68 (71%)
Furst 2003	STAR	csDMARD-IR	DMARD	NR	NR	318	237 (75%)	Low
			ADA + DMARD	NR	NR	318	243 (76%)
Klareskog 2004	TEMPO	csDMARD-IR	MTX	NR	NR	219	164 (75%)	Low
			ETN + MTX	NR	NR	227	177 (78%)
Genovese 2008	TOWARD	csDMARD-IR	DMARD	NR	NR	413	311 (75%)	Low
			TCZ + DMARD	NR	NR	803	624 (78%)
3. TNFi-IR
Genovese 2005	ATTAIN	TNFi-IR	DMARD	NR	NR	127	97 (76%)	Low
			ABT + DMARD	NR	NR	241	189 (78%)
Smolen 2009	GO-AFTER	TNFi-IR	DMARD	NR	NR	146	108 (74%)	Low
			GLM + DMARD	NR	NR	143	104 (73%)
Emery 2008	RADIATE	TNFi-IR	MTX	NR	NR	160	120 (75%)	Low
			TCZ + MTX	NR	NR	175	139 (79%)
B. RCTs of bDMARD monotherapy
Kaneko 2016	SURPRISE	MTX-IR	TCZ	NR	NR	102	82 (80%)	Highb
			TCZ + MTX	NR	NR	105	78 (74%)
Jones 2010	AMBITION	csDMARD-IR	MTX	NR	NR	284	212 (75%)	Low
			TCZ	NR	NR	288	214 (74%)
Takeuchi 2013		csDMARD-IR	MTX	NR	NR	176	138 (78%)	Low
			ETN	NR	NR	182	147 (81%)
C. Head-to-head RCTs of bDMARDs
Gabay 2013	ADACTA	MTX-IR	ADA	162	115 (71%)	162	119 (73%)	Low
			TCZ	163	125 (77%)	163	121 (74%)
Weinblatt 2013	AMPLE	MTX-IR	ABT	251	185 (74%)	251	204 (81%)	Highc
			ADA	257	203 (79%)	258	223 (86%)

a

Open-label design.

b

Open-label design.

c

Assessor-blinded but patients were not blinded.

ABT: abatacept; ACPA: anticitrullinated protein antibody; ADA: adalimumab; csDMARD: conventional synthetic disease-modifying antirheumatic drug; CZP: certolizumab pegol; ETN: etanercept; GLM: golimumab; IFX: infliximab; IR: inadequate response; MTX: methotrexate; NR: not recorded; RF: rheumatoid factor; RTX: rituximab; SSZ: sulfasalazine; TCZ: tocilizumab; TNFi: tumor necrosis factor inhibitor. References are summarized them in the Supplementary material (available at Rheumatology online).

RCTs of biological DMARDs with conventional synthetic DMARDs vs conventional synthetic DMARDs

A total of 23 of the RCTs compared bDMARDs+csDMARDs with csDMARDs (±placebo) and included (i) MTX-naive or csDMARD-naive patients (n = 6 trials; all TNFi trials); (ii) MTX-IR or csDMARD-IR (n = 14 trials; TNFi n = 9, abatacept n = 1, RTX n = 2, tocilizumab n = 2); or (iii) TNFi-IR (n = 3 trials; TNFi n = 1, abatacept n = 1, tocilizumab n = 1) RA patients. As shown in Fig. 1A, among the retrieved six trials in MTX-naive or csDMARD-naive patients, four RCTs could be included in the meta-analysis and the pooled RRR for the 6-month ACR20 response comparing ACPA-positive to ACPA-negative in the bDMARD-group vs the non-bDMARD-group was 0.95 (95% CI 0.81–1.11) (Fig. 1A). For RF-positive vs RF-negative patients from three RCTs, the pooled RRR for the 6-month ACR20 response was 1.02 (95% CI 0.88–1.18) (Fig. 1B). Other outcomes followed a similar pattern, with no difference between seropositive and seronegative patients (Table 2).

Figure 1.

Forest plots for the Relative Risk Ratio (RRR) for ACR20 at 6 months for csDMARD-naive RA patients. When the RRR is 1, the efficacy is equal between autoantibody-positive and autoantibody-negative RA patients. A value >1 indicates a higher efficacy in autoantibody-positive RA patients. (A) ACPA(+)/ACPA(-) in bDMARD+csDMARD vs ACPA(+)/ACPA(-) in csDMARD. (B) RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD

Table 2.

Summary of efficacy outcome measures at 6 months

Outcomes	Number of trials	Relative risk ratios/Differences of differences (95% CI)	I2
1. MTX-naive or csDMARD-naive
ACPA(+)/ACPA(-) in bDMARD+csDMARD vs ACPA(+)/ACPA(-) in csDMARD
		Relative risk ratios (95%CI)
ACR20	n = 4	0.95 (0.81, 1.11)	0%
ACR50	n = 4	0.95 (0.78, 1.15)	0%
ACR70	n = 4	0.79 (0.55, 1.13)	0%
DAS28-ESR remission	n = 6	0.95 (0.71, 1.27)	0%
DAS28-CRP remission	n = 5	0.80 (0.57, 1.12)	18%
HAQ-DI remission	n = 6	1.07 (0.90, 1.27)	20%
		Difference of differences (95% CI)
Delta DAS28-ESR	n = 6	0.04 (−0.32, 0.39)	0%
Delta DAS28-CRP	n = 5	−0.02 (−0.38, 0.35)	0%
Delta HAQ-DI	n = 6	−0.09 (−0.23, 0.05)	0%
RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD
		Relative risk ratios (95%CI)
ACR20	n = 3	1.02 (0.88, 1.18)	0%
ACR50	n = 3	0.94 (0.76, 1.16)	0%
ACR70	n = 3	0.80 (0.57, 1.12)	0%
DAS28-ESR remission	n = 5	1.03 (0.86, 1.25)	0%
DAS28-CRP remission	n = 4	0.96 (0.74, 1.24)	0%
HAQ-DI remission	n = 5	1.08 (0.85, 1.36)	33%
		Difference of differences (95% CI)
Delta DAS28-ESR	n = 5	−0.05 (−0.44, 0.35)	0%
Delta DAS28-CRP	n = 4	−0.01 (−0.41, 0.38)	0%
Delta HAQ-DI	n = 5	−0.12 (−0.27, 0.04)	0%
2. MTX-IR or csDMARD-IR
RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD
		Relative risk ratios (95%CI)
ACR20	n = 13	1.09 (0.90, 1.32)	15%
ACR50	n = 13	1.13 (0.78, 1.65)	24%
ACR70	n = 12	1.79 (1.01, 3.16)	11%
DAS28-ESR remission	n = 8	1.22 (0.41, 3.60)	24%
DAS28-CRP remission	n = 8	1.28 (0.69, 2.39)	0%
HAQ-DI remission	n = 8	1.00 (0.84, 1.18)	0%
		Difference of differences (95% CI)
Delta DAS28-ESR	n = 8	−0.08 (−0.33, 0.17)	0%
Delta DAS28-CRP	n = 8	−0.18 (−0.42, 0.06)	0%
Delta HAQ-DI	n = 8	−0.06 (−0.16, 0.04)	0%
3. TNFi-IR
RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD
		Relative risk ratios (95%CI)
ACR20	n = 3	2.28 (1.31, 3.95)	0%
ACR50	n = 3	4.50 (1.35, 15.02)	0%
ACR70	n = 3	3.69 (0.42, 32.62)	0%
DAS28-ESR remission	n = 2	6.56 (0.69, 62.00)	0%
DAS28-CRP remission	n = 2	4.08 (0.38, 44.14)	0%
HAQ-DI remission	n = 2	0.88 (0.72, 1.08)	0%
		Difference of differences (95% CI)
Delta DAS28-ESR	n = 2	−0.57 (−1.04, −0.10)	1%
Delta DAS28-CRP	n = 2	−0.45 (−0.95, 0.05)	29%
Delta HAQ-DI	n = 2	−0.04 (−0.21, 0.12)	0%

Bold: estimates that are statistically significant, i.e. P < 0.05. bDMARD: biological synthetic disease-modifying antirheumatic drug; csDMARD: conventional synthetic disease-modifying antirheumatic drug.

Fig. 2A depicts the data for the MTX-IR or csDMARD-IR group. Among the retrieved 14 trials, 13 RCTs had data available at 24 weeks and were included in the meta-analysis. The pooled RRR for the 6-month ACR20 response comparing RF-positive to RF-negative in bDMARD vs non-bDMARD-group was 1.09 (95% CI 0.90–1.32). Other outcome measures confirmed that there was no association between RF and the efficacy of bDMARD treatment in csDMARD-IR patients as shown in Table 2, apart from perhaps ACR70-responses, which appeared to be more readily achieved by RF-positive patients. RCTs included in the MTX-IR or csDMARD-IR group measured only RF and lacked data on ACPA, so a pooled RRR comparing ACPA-positive to ACPA-negative could not be calculated.

Figure 2.

(A) Forest plot for the Relative Risk Ratio for ACR20 at 6 months for csDMARD-IR RA patients comparing RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD. (B) Forest plot for the Relative Risk Ratio for ACR20 at 6 months for TNFi-IR RA patients comparing RF(+)/RF(-) in bDMARD+csDMARD vs RF(+)/RF(-) in csDMARD

In the analysis in the TNFi-IR group, including three trials, the RRR were compatible with greater efficacy of bDMARDs in RF-positive patients compared with RF-negative patients with the RRRs for the ACR20 and 50 responses being 2.28 (95% CI 1.31–3.95) and 4.50 (95% CI 1.35–15.02), respectively (Fig. 2B, Table 2). The same effect was also seen in the decrease in disease activity with delta DAS28-ESR being -0.57 (95% CI -1.04 to -0.10) in RF-positive vs RF-negative patients. Other outcomes showed a similar effect, although confidence intervals were large leading to non-significant differences (Table 2).

Regarding the outcome of radiographic damage, for the majority of the trial data accessed through international trial data repositories, we obtained data in different formats for baseline and 6 months, hampering the calculation of progression, our outcome of interest. In the few trials with data available on infliximab (IDEA) and certolizumab pegol (RAPID1, RAPID2), radiographic progression at 6 months was not different between autoantibody-positive and autoantibody-negative patients (Supplementary Tables S1.6 and S2.6, available at Rheumatology online).

RCTs of biological DMARD monotherapy

We found one trial comparing bDMARD+MTX combination therapy vs bDMARD monotherapy and two RCTs with csDMARD-IR patients in which bDMARD monotherapy was compared with MTX monotherapy. The results are listed in Table 3. The first trial comparing tocilizumab+MTX with tocilizumab reported a higher efficacy in RF-positive compared with RF-negative patients in terms of ACR20 and 70 but not in the remaining outcomes, for which there was no difference between the groups. The other trials did not show any differences according to the serological status.

Table 3.

Summary of the experimental intervention, comparator and outcomes of interest at 6 months

Outcomes	Intervention	Comparator	Relative risk ratios (95% CI)
RCTs of bDMARD monotherapy
RF(+)/RF(-) in intervention treatment vs RF(+)/RF(-) in comparator treatment
Kaneko 2016 (SURPRISE)
ACR20	Tocilizumab + MTX	Tocilizumab	1.62 (1.06–2.47)
ACR50			1.75 (0.94–3.26)
ACR70			3.05 (1.12–8.27)
DAS28-ESR remission			0.97 (0.57–1.65)
DAS28-CRP remission			1.09 (0.74–1.60)
HAQ-DI remission			0.72 (0.38–1.35)
Jones 2010 (AMBITION)
ACR20	Tocilizumab	MTX	0.83 (0.58–1.19)
ACR50			1.07 (0.63–1.81)
ACR70			0.73 (0.30–1.80)
DAS28-ESR remission			NA
DAS28-CRP remission			NA
HAQ-DI remission			NA
Takeuchi 2013
ACR20	Etanercept	MTX	1.01 (0.64–1.59)
ACR50			1.12 (0.56–2.24)
ACR70			0.73 (0.22–2.40)
DAS28-ESR remission			0.63 (0.21–1.85)
DAS28-CRP remission			1.26 (0.56–2.81)
HAQ-DI remission			0.96 (0.60–1.54)
Head to head RCTs of bDMARDs
ACPA(+)/ACPA(-) in intervention treatment vs ACPA(+)/ACPA(-) in comparator treatment
Gabay 2013 (ADACTA)
ACR20	Tocilizumab	Adalimumab	1.39 (0.91–2.14)
ACR50			1.82 (0.88–3.77)
ACR70			1.09 (0.38–3.15)
DAS28-ESR remission			NA
DAS28-CRP remission			NA
HAQ-DI remission			NA
Weinblatt 2013 (AMPLE)
ACR20	Abatacept + MTX	Adalimumab + MTX	0.76 (0.54–1.06)
ACR50			1.09 (0.60–1.96)
ACR70			1.08 (0.42–2.78)
DAS28-ESR remission			NA
DAS28-CRP remission			0.95 (0.48–1.86)
HAQ-DI remission			0.87 (0.61–1.24)
RF(+)/RF(-) in intervention treatment vs RF(+)/RF(-) in comparator treatment
Gabay 2013 (ADACTA)
ACR20	Tocilizumab	Adalimumab	1.52 (0.98–2.35)
ACR50			2.14 (1.05–4.36)
ACR70			1.13 (0.37–3.44)
DAS28-ESR remission			NA
DAS28-CRP remission			NA
HAQ-DI remission			NA
Weinblatt 2013 (AMPLE)
ACR20	Abatacept + MTX	Adalimumab + MTX	0.94 (0.63–1.42)
ACR50			1.11 (0.51–2.41)
ACR70			1.34 (0.45–3.99)
DAS28-ESR remission			NA
DAS28-CRP remission			1.93 (0.85–4.39)
HAQ-DI remission			0.99 (0.66–1.50)

Bold: estimates that are statistically significant, i.e. P < 0.05. bDMARD: biological synthetic disease-modifying antirheumatic drug; csDMARD: conventional synthetic disease-modifying antirheumatic drug; MTX: methotrexate; NA: not applicable.

Head-to-head RCTs of biological DMARDs

Two trials comparing two bDMARDs head-to-head were analysed (Table 3). In the ADACTA trial, the efficacy of tocilizumab monotherapy compared with adalimumab monotherapy was assessed. ACR50 was higher in RF-positive compared with RF-negative patients when comparing tocilizumab vs adalimumab, but this difference was not found when comparing ACPA-positive to ACPA-negative patients, nor for the remaining outcomes. AMPLE was an RCT comparing abatacept and adalimumab in bDMARD-naive RA patients with background MTX, in which no differences were seen in the outcomes comparing two groups (RF+ vs RF-, ACPA+ vs ACPA-).

The impact of the mode of action

To investigate whether the association between seropositivity and treatment response might differ depending on the drug’s mode of action or on the individual drug, we performed sensitivity analyses for TNFi vs non-TNFi bDMARDs, as well as for the individual agents. The results indicated comparable efficacy between TNFi and non-TNFi for both RF-positive and RF-negative patients with RA (Table 4). Furthermore, individual analysis of each bDMARD revealed no statistically significant difference in efficacy between RF-positive and RF-negative patients (Table 4). Although occasionally an outcome measure achieved statistical significance according to the P <0.05 threshold, there was no consistent signal.

Table 4.

Summary of sensitivity analyses at 6 months^a

Outcomes	Intervention	Comparator	Relative risk ratios (95% CI)
1. Mode of action
RF(+)/RF(-) in intervention treatment vs RF(+)/RF(-) in comparator treatment
TNFi
ACR20	TNFi+csDMARD	csDMARD	1.03 (0.92–1.15)
ACR50			0.99 (0.77–1.27)
ACR70			0.87 (0.64–1.17)
DAS28-ESR remission			1.10 (0.86–1.39)
DAS28-CRP remission			1.01 (0.79–1.28)
HAQ-DI remission			1.03 (0.90–1.17)
Non-TNFi
ACR20	Non–TNFi+csDMARD	csDMARD	1.31 (0.99–1.75)
ACR50			1.38 (0.88-2.16)
ACR70			2.85 (1.41–5.76)
DAS28-ESR remission			1.29 (0.21–8.04)
DAS28-CRP remission			1.36 (0.33–5.57)
HAQ-DI remission			0.78 (0.52–1.16)
2. Individual bDMARDs
RF(+)/RF(-) in intervention treatment vs RF(+)/RF(-) in comparator treatment
Infliximab
ACR20	Infliximab+csDMARD	csDMARD	1.02 (0.88–1.18)
ACR50			0.94 (0.75–1.16)
ACR70			0.83 (0.58–1.19)
DAS28-ESR remission			1.06 (0.86–1.30)
DAS28-CRP remission			0.92 (0.68–1.23)
HAQ-DI remission			1.05 (0.58–1.87)
Etanercept
ACR20	Etanercept+csDMARD	csDMARD	0.62 (0.22–1.77)
ACR50			0.55 (0.31–0.96)
ACR70			0.98 (0.41–2.34)
DAS28-ESR remission			1.73 (0.63–4.79)
DAS28-CRP remission			1.27 (0.76–2.12)
HAQ-DI remission			1.04 (0.85–1.28)
Adalimumab
ACR20	Adalimumab+csDMARD	csDMARD	1.30 (0.97–1.75)
ACR50			1.32 (0.80–2.15)
ACR70			0.66 (0.27–1.58)
DAS28-ESR remission			0.74 (0.39–1.42)
DAS28-CRP remission			0.63 (0.25–1.61)
HAQ-DI remission			1.03 (0.75–1.42)
Certolizumab-pegol
ACR20	Certolizumab-pegol+csDMARD	csDMARD	1.13 (0.65–1.95)
ACR50			1.35 (0.54–3.33)
ACR70			2.81 (0.36–21.97)
DAS28-ESR remission			0.59 (0.12–2.95)
DAS28-CRP remission			2.09 (0.64–6.82)
HAQ-DI remission			1.04 (0.74–1.46)
Golimumab
ACR20	Golimumab+csDMARD	csDMARD	1.82 (0.61–5.43)
ACR50			9.14 (0.86–97.48)
ACR70			5.33 (0.32–88.94)
DAS28-ESR remission			8.74 (0.67–113.87)
DAS28-CRP remission			5.80 (0.35–96.37)
HAQ-DI remission			0.90 (0.73–1.10)
Abatacept
ACR20	Abatacept+csDMARD	csDMARD	1.54 (0.55–4.35)
ACR50			1.42 (0.13–15.62)
ACR70			1.49 (0.35–6.32)
DAS28-ESR remission			1.53 (0.06–36.76)
DAS28-CRP remission			0.64 (0.03–13.00)
HAQ-DI remission			0.79 (0.52–1.20)
Rituximab
ACR20	Rituximab+csDMARD	csDMARD	1.06 (0.61–1.82)
ACR50			1.49 (0.59–3.75)
ACR70			1.83 (0.46–7.34)
DAS28-ESR remission			1.18 (0.13–11.09)
DAS28-CRP remission			1.67 (0.34–8.29)
HAQ-DI remission			0.65 (0.14–3.02)
Tocilizumab
ACR20	Tocilizumab+csDMARD	csDMARD	1.39 (0.92–2.10)
ACR50			1.44 (0.80–2.59)
ACR70			4.83 (1.80–12.94)
DAS28-ESR remission			NA
DAS28-CRP remission			NA
HAQ-DI remission			NA

a

Analyses with all eligible studies, regardless of previous (b)DMARD exposure. Bold: estimates that are statistically significant, i.e. P < 0.05.

bDMARD: biological synthetic disease-modifying antirheumatic drug; csDMARD: conventional synthetic disease-modifying antirheumatic drug.

Discussion

In this study, we report a meta-analysis based on an SLR including solely RCT data to investigate whether the efficacy of bDMARDs differs in autoantibody-positive compared with autoantibody-negative RA. In the csDMARD-naive/csDMARD-IR patients, positive autoantibodies are not associated with a better response to bDMARDs. In TNFi-IR patients, the meta-analysis showed that bDMARDs may be more effective in the autoantibody-positive patients, but this finding should be interpreted with caution due to the limited number of trials and inconsistent findings across the outcomes assessed.

How do these results align with previous findings? Previous RCT-publications have not provided conclusive evidence on whether treatment responses to bDMARDs differ in autoantibody-positive and -negative patients. For RTX, there have been several reports which together do not produce a clear picture [29, 30]. A conference abstract reported that in the RTX-trials MIRROR and SERENE, consisting of MTX-IR patients, autoantibody-positive (either or both RF/ACPA) patients were more likely to achieve ACR20/50/70 response and DAS28 low disease activity [31]. In a later publication onTNFi-IR patients, positivity for both RF and ACPA autoantibodies (IgG anti-CCP, IgM-RF, IgG-RF and IgA-RF) was found to be associated with a better treatment response, but only in the presence of high concentrations of CRP [32]. The results of a meta-analysis of previous RTX-RCTs, which included patients with various pre-treatments (csDMARD-naive, csDMARD-IR and TNFi-IR), are consistent with our analysis, showing that autoantibody-positive patients only responded better to RTX in the TNFi-IR subgroup [13].

Regarding other bDMARDs, data from RCTs are also limited. The AMPLE trial previously indicated differing response patterns to abatacept and adalimumab based on ACPA concentration [33]. However, when comparing the responses in ACPA+ vs ACPA- between the abatacept and adalimumab groups, no differences were found, as indicated by a RRR ≈1 (calculated in our study). As ACPA concentrations were not available for the remaining trials included in this SLR, we decided to restrict our analyses to the comparison between autoantibody-positive and autoantibody-negative (RF+ vs RF-, ACPA+ vs ACPA-) patients only. This potential difference in response across drugs based on the ACPA concentration remains to be further investigated.

Observational studies have more frequently reported about the association between autoantibody status and response to bDMARDs. A previous meta-analysis described the effects of autoantibody positivity on response to TNFi treatment [14]. A total of 14 studies examining RA populations with various pre-treatments (csDMARD-IR and TNFi-IR) were included, 13 of which were observational studies and only one was an RCT (the GO-FORWARD trial). Although our current meta-analysis excluded GO-FORWARD trial (due to <20% seronegative patients), the overall findings remain consistent: RF- nor ACPA-status do not impact the clinical efficacy of TNFi treatment in RA. The evidence regarding a higher efficacy of RTX and abatacept in seropositive patients from observational studies is primarily based on patients with longer disease duration including TNFi-IR patients [5–12]. Thus, it is important to note that although our findings are very much in line with these previous reports, they nonetheless cast a new light on the belief, widely held in daily clinical practice, that ‘in the real world’ RTX and abatacept would be more efficacious in seropositive patients. This notion only holds true in the sense that many patients first receive a TNFi, and only upon failing TNFi-treatment are prescribed RTX or abatacept, when chances that seronegative patients will respond may already have diminished, as described below.

Several factors (apart from a true difference between seropositive and seronegative patients) likely play a role in explaining these results in TNFi-IR patients. First, the possibility of misdiagnosis in seronegative RA patients. Despite the fact that all patients fulfilled RA classification criteria, a degree of uncertainty always remains in the diagnosis of seronegative RA. Moreover, the fact that these patients had not responded to previous treatments, including TNFi, raises the question whether their disease was at all amenable to anti-inflammatory treatment. From this perspective, the fact that seronegative TNFi-IR patients responded less well than seropositive TNFi-IR was perhaps not surprising. A second important factor may be inherent to the current outcome measures such ACR- and DAS-response, which can be influenced by patient and physician expectation, even in double-blind RCTs. In a recent elegant study [34], absolute treatment effects such as ACR-response were consistently found to be larger in bDMARD head-to-head trials than in placebo-controlled trials. This suggests that, when both patients and physicians know that a drug is being administered (as in head-to-head trials), the treatment response may become larger due to psychological mechanisms. This form of expectation bias may therefore have influenced the results and cannot easily be prevented. Finally, assuming that misdiagnosis and expectation bias could be overcome, a possible explanation for our findings could also be that autoantibody-negative RA patients have very heterogeneous underlying pathophysiological mechanisms. More research is needed to disentangle the entity currently known as ‘seronegative RA’.

Taking all these factors into account, our interpretation of the results of our meta-analysis is that the efficacy of bDMARDs is in general comparable in seropositive and seronegative patients. The small number of included trials in the TNFi-IR group (the only one in which potential differences between groups were found), the lack of consistency across outcomes, the large confidence intervals, together with the fact that all sensitivity analyses showed no differences between seropositive and seronegative patents (i.e., between bDMARDs grouped by mode of action, and by individual drug analysed) support this interpretation.

There are several limitations to our study, mostly pertaining to the number of studies. To enable a comparison of autoantibody-positive and autoantibody-negative patients within each study, we only included RCTs in which autoantibody-positive patients constituted <80% of the total study population. This resulted in a limited number of studies, particularly in the bDMARD-IR category. This cutoff of 80% is arbitrary, but we wanted to ensure that we had a minimal representation of seronegative patients in order to compare the efficacy of bDMARDs between both groups. Furthermore, we were unable to analyse some of the outcomes, such as radiographic progression due to lack of data. Nonetheless, the findings for the different outcomes which were available (ACR20, 50 or 70 and DAS-based outcomes) showed the same effect arguing in favour of robustness of our findings. In addition, data on ACPA were considerably scarcer than data on RF, and some analyses could therefore only be performed for RF. These limitations of the data set prohibited separate sub-analyses of ACPA- and RF-double-positive vs single-positive vs double-negative patients. Moreover, this study focused on bDMARD and nowadays it would be good to extend the analysis to Janus kinase inhibitors (JAKi) as well. However, the project was initiated in 2016, when there were very few trials on JAKi published, and it took a very long time to obtain access to trial data. Starting attempts to obtain JAKi later would have extended the project even further and was therefore not feasible. Along the same line, only studies published before 2016 could be included, which is a limitation also for the exclusion of studies on bDMARDs published since then [35–37]. However, the data we needed was mainly unpublished and the process of access to trial data is extremely long. Therefore, this meta-analysis based on an SLR also has the added value of publishing the outcomes of the included trials stratified by the autoantibody status (Supplementary Material, available at Rheumatology online), which can be further used by researchers.

Strengths of our study include the unbiased approach of collecting and analysing all data available on this topic from RCTs. This includes unpublished data from existing RCTs, which is even more valuable as these data are not only meta-analysed, but now also made available at the individual trial level in the Supplementary Material, available at Rheumatology online. This provides a very important additional perspective to the data reported from observational studies thus far, uncovering essential and hitherto underappreciated differences, such as the divergent patterns observed between TNFi-IR patients and less treatment-resistant groups.

To summarize, in this meta-analysis of data from RCTs, we answer a question that has hovered in the air, namely whether the presence of RF/ACPA affects the efficacy of bDMARDs and in general it is comparable, regardless of the patient population, mechanism of action or individual drug used.

Supplementary material

Supplementary material is available at Rheumatology online.

Data availability

Data are available on reasonable request.

Contribution statement

K.T.-M., S.B., J.L.N., D.vdH., R.L., S.R. and D.vdW. participated in the design of the study. All authors participated in the development of the project, the interpretation of the data and the manuscript preparation, and approved the current version of the manuscript.

Funding

The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

Disclosure statement: K.T.-M. reports speaker fees from Daiichi Sankyo, Gilead and Mitsubishi Tanabe. Y.K. received grants from AbbVie, Asahi Kasei Pharma Corp., AYUMI Pharmaceutical Corp., Chugai Pharmaceutical Co., Ltd, Eisai Co., Ltd, Mitsubishi-Tanabe Pharma Corp., Lt., and Boehringer Ingelheim, and lecture fees from AbbVie, Asahi Kasei Pharma Corp., Astellas, Bristol-Myers Squibb, Chugai Pharmaceutical Co., Ltd, Daiichi Sankyo Co. Ltd, Eisai Co., Ltd, Eli Lilly Japan, Gilead Sciences, Inc., Janssen Pharmaceutical K.K., Mitsubishi-Tanabe Pharma Corp., Pfizer Japan Inc., Boehringer Ingelheim, and Novartis Pharma. F.B. reports research grants from AbbVie, Prophylix and consulting fees from Amgen, Biotest, Boehringer, Bristol Myers Squibb, Celgene, Chugai, Genzyme, Janssen, Lilly, MSD, Novartis, Pfizer, Roche, Sandoz, Sanofi, UCB and speaker fees from AbbVie, Amgen, Biotest, Boehringer, Bristol Myers Squibb, Celgene, Chugai, Genzyme, Janssen, Lilly, MSD, Novartis, Pfizer, Roche, Sandoz, Sanofi, UCB. S.S. is an employee of deCODE genetics. D.vdH. received consulting fees from AbbVie, ArgenX, Bayer, BMS, Galapagos, Gilead, Glaxo-Smith-Kline, Janssen, Lilly, Novartis, Pfizer, Takeda, UCB Pharma and director of Imaging Rheumatology BV. R.L. reports research grants from FOREUM, AbbVie, UCB, Pfizer, Novartis and consulting fees from AbbVie, Eli-Lilly, Galapagos, Jansen pharma, Pfizer, Novartis, UCB and lecture fees from AbbVie, Eli-Lilly, Galapagos, Jansen pharma, Pfizer, Novartis, UCB and director of Joint Imaging BV. S.R. reports research grants from AbbVie, Galapagos, MSD, Novartis, Pfizer, UCB and consulting fees from AbbVie, Eli Lilly, MSD, Novartis, Pfizer, Sanofi, UCB and lecture fees from AbbVie, Eli-Lilly, Pfizer, Novartis, UCB. D.vdW. reports consulting and lecture fees from Galapagos Biopharma.

Acknowledgements

We obtained published and unpublished investigator-initiated trial data through contact with the authors. We thank all the members of the following study groups: COMET, EMPIRE, HIT HARD, IDEA, NEO-RACo, SWEFOT, AIM, Choy 2012, Kay 2008, Kim 2007, OPTION, RAPID1, RAPID2, SERENE, AMARA, CERTAIN, Combe 2006, STAR, TEMPO, TOWARD, ATTAIN, GOAFTER, RADIATE, SURPRISE, AMBITION, Takeuchi 2013, ADACTA, AMPLE study groups. We would also like to acknowledge the contribution from Karen Hambardzumyan (SWEFOT) and Michaela Koeh (AMARA).

This manuscript is based on research using data from Pfizer Inc. and AbbVie Inc. that has been made available through Vivli, Inc. Vivli has not contributed to or approved, and is not in any way responsible for, the contents of this manuscript. This study, carried out under YODA Project #2017–2381, used data obtained from the Yale University Open Data Access Project, which has an agreement with JANSSEN RESEARCH & DEVELOPMENT, L.L.C. The interpretation and reporting of research using this data are solely the responsibility of the authors and does not necessarily represent the official views of the Yale University Open Data Access Project or JANSSEN RESEARCH & DEVELOPMENT, L.L.C. The datasets of Roche (CSDR Research Proposal 5808) generated and analysed during this study are available in anonymized format upon reasonable request via the CSDR platform (http://www.clinicalstudydatarequest.com). This manuscript is based in part on data from the UCB, BMS, and JANSSEN RESEARCH & DEVELOPMENT, L.L.C.

Author notes

S.R. and D.v.d.W. contributed equally.