Risk of flare after tapering or withdrawal of biologic/targeted synthetic disease-modifying anti-rheumatic drugs in patients with rheumatoid arthritis or axial spondyloarthritis: a systematic review and meta-analysis

Abstract

Objective

To evaluate flare risk when tapering or withdrawing biologic or targeted synthetic DMARDs (bDMARDs or tsDMARDs) compared with continuation in patients with inflammatory arthritis in sustained remission or with low disease activity.

Methods

Articles were identified in the Cochrane Library, PubMed, Embase and Web of Science. Eligible trials were randomized controlled trials comparing tapering and/or withdrawal of bDMARDs and/or tsDMARDs with the standard dose in inflammatory arthritis. Random effects meta-analysis was performed with risk ratio (RR) or Peto’s odds ratio (POR) for sparse events and 95% CI.

Results

The meta-analysis comprised 22 trials: 11 assessed tapering and 7 addressed withdrawal (4 assessed both). Only trials with an RA or axial SpA (axSpA) population were identified. An increased flare risk was demonstrated when b-/tsDMARD tapering was compared with continuation [RR 1.45 (95% CI 1.19, 1.77), I² = 42.5%] and potentially increased for persistent flare [POR 1.56 (95% CI 0.97, 2.52), I² = 0%]. Comparing TNF inhibitor (TNFi) withdrawal with continuation, a highly increased flare risk [RR 2.28 (95% CI 1.78, 2.93), I² = 78%] and increased odds of persistent flare [POR 3.41 (95% CI 1.91, 6.09), I² = 49%] were observed. No clear difference in flare risk between RA or axSpA was observed.

Conclusion

A high risk for flare and persistent flare was demonstrated for TNFi withdrawal, whereas an increased risk for flare but not for persistent flare was observed for b-/tsDMARD tapering. Thus tapering seems to be the more favourable approach.

Registration

PROSPERO (CRD42019136905).

biologic therapies, RA, spondylarthropathies (including psoriatic arthritis), systematic review, meta-analysis

Rheumatology key messages

• Tapering b-/tsDMARDs in RA and axSpA results in an increased risk of flare but not persistent flare.

• Withdrawing TNFis in RA and axSpA results in both an increased risk of flare and persistent flare.

• Tapering seems to be the more favourable choice for a dose reduction strategy in RA and axSpA.

Introduction

In recent years, a topic of great interest has been whether biologic or targeted synthetic DMARDs (bDMARDs or tsDMARDs, respectively) should be continued indefinitely in patients with inflammatory arthritis (IA) [i.e. RA, PsA and axial SpA (axSpA)] in sustained remission or low disease activity (LDA). Lifelong continuation of the standard dose can potentially lead to overtreatment, thereby putting the individual patient at an unnecessary risk of adverse events (e.g. serious infections) [1]. In addition to cost savings (as b-/tsDMARDs are expensive), dose reduction by interval prolonging or complete drug withdrawal would result in fewer visits to the outpatient clinic for intravenous infusions or for medicine pickup. Previous research has shown that patient quality of life and patient satisfaction improve when the disease is controlled with fewer drug doses, as patients feel less dependent on their medication [2]. A recent qualitative study in RA observed a fear of arthritis flare and joint damage when discussing tapering of bDMARDs; however, since tapering would minimize the inconvenience of taking bDMARDs regularly, patients were willing to taper but not to completely withdraw [3].

Currently, several international guidelines recommend tapering of b-/tsDMARDs in patients with IA in sustained remission, whereas complete withdrawal is generally not recommended due to an increased flare risk [4–6]. In the existing literature, various flare criteria have been used to evaluate patients with IA, as validated flare criteria were developed in 2013 for RA [28-joint DAS with CRP (DAS28-CRP) increase >1.2 or >0.6 if current DAS28-CRP ≥3.2] [7] and in 2018 for axSpA (ASDAS increase ≥0.9) [8], whereas none exist for PsA. The heterogeneity in flare criteria must be taken into consideration when comparing trial results.

Previously, evidence from meta-analyses found tapering of bDMARDs in patients with RA in sustained remission/LDA to be comparable to continuation of the standard dose [9–11], whereas complete withdrawal was potentially inferior [9, 10, 12]. To our knowledge, no meta-analyses have assessed flare risk when tapering or withdrawing b-/tsDMARDs compared with continuation of the standard dose across all main forms of IA (spanning RA, axSpA and PsA). However, recent reviews found tapering but not withdrawal of bDMARDs to be comparable to continuation of the standard dose in axSpA and PsA [13–15].

This systematic literature review (SLR) aims to assess the risk for flare in patients with IA in sustained remission/LDA randomized to either tapering or withdrawal of various b-/tsDMARDs compared with continuation of treatment.

Materials and methods

Patient and public involvement

The conduct of this SLR was discussed with and acknowledged by patients treated with b-/tsDMARDs in sustained remission/LDA from the Department of Rheumatology, Aalborg University Hospital. Results were discussed at an informal meeting prior to submission.

Protocol and registration

This SLR was performed in accordance with the pre-specified protocol (Supplementary Data S1, available at Rheumatology online) as recommended in the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) protocol guideline [16, 17] and registered at the International Prospective Register of Systematic Reviews (PROSPERO) on 27 May 2019 (CRD42019136905). Furthermore, the review was performed and reported in accordance with the recommendations of the Methodological Expectations of Cochrane Intervention Reviews [18] and the PRISMA statement [19].

Search strategy

A systematic literature search was performed to identify randomized controlled trials (RCTs) comparing tapering or withdrawal of b-/tsDMARDs with continuation of the standard dose in adults with RA, PsA or axSpA. The minimum required follow-up period was 24 weeks. The search strategy (Supplementary Data S2, available at Rheumatology online) was developed in collaboration with a trained medical research librarian (C.S.). On 13–14 June 2019, the Cochrane Library, PubMed, Embase and Web of Science were searched. The search period was from the inception of the database until the search date (rerun on 11 January 2021). Moreover, congress abstracts from EULAR and ACR after 1 January 2017 and reference lists of included articles/excluded reviews were reviewed for eligibility.

Study selection process

All identified records were uploaded in Endnote to remove duplicates, then imported to the Covidence Technology Platform (Covidence, Melbourne, VIC, Australia). The screening process was divided into three phases; all assessors were trained with a calibration exercise to ensure reviewer consistency. In phase 1, L.U., W.K.H.D., C.H.L. and S.S.A. assessed titles and abstracts for relevance (yes, no or unclear). Thereafter, L.U. and S.K. independently evaluated the abstracts for eligibility. In phase 3, L.U. and S.K. reviewed full-text articles for inclusion. Disagreements were resolved by discussion and, if necessary, by consultation with a third independent assessor (L.D.).

Data collection process

A unique data extraction sheet was developed in Covidence based on the Cochrane Systematic Review data collection form. Data extraction was done independently by L.U. and S.K., who resolved any disagreements by discussion and, if necessary, by consultation with a third assessor (L.D. or R.C.). Multiple reports of the same study were collated as recommended by Cochrane [18]. Extracted data included journal, publication year, first author, study design, eligibility criteria, disease duration, duration of remission/LDA before inclusion, assessed b-/tsDMARDs, intervention and comparator details, risk of flare and persistent flare, risk of adverse events such as serious infections and identification of possible prognostic factors.

Study risk of bias assessment

For each eligible trial, the internal validity was evaluated using the revised Cochrane Risk of Bias tool for randomized trials (RoB 2.0) template [20], thus the following six domains were assessed: risk of bias arising from the randomization process, risk of bias due to deviations from the intended interventions, missing outcome data, risk of bias in measurement of the outcome, risk of bias in selection of the reported result and the overall risk of bias. The domains were assessed as either ‘low risk of bias’, ‘some concerns’ or ‘high risk of bias’. L.U. performed the risk of bias assessment, then S.K. checked the assessment for agreement. Differences were resolved by discussion and consensus. Corresponding authors of 16 studies were contacted by e-mail (a maximum of three e-mails) to gain access to the trial protocol and statistical analysis plan for the risk of bias assessment. Seven authors did not respond to the e-mails.

Effect measures

The primary statistical effect measure was the risk ratio (RR) with 95% CI for flares [21]. Secondary outcomes were due to anticipated sparse events analysed with Peto’s odds ratio (POR) [22]: persistent flare (i.e. flares with no improvement despite dose escalation or glucocorticoids) and serious adverse events (SAEs), including serious infections.

Synthesis methods

Meta-analyses were performed based on restricted maximum likelihood mixed-effects models with trials applied as random effects. I² statistics were used to evaluate inconsistencies among studies and prediction intervals were added to facilitate interpretation of the impact of between-study heterogeneity [23, 24]. Two trials had more than one tapering strategy [25, 26], therefore data from the two tapering groups in each study were pooled. For the purpose of sensitivity, stratified meta-analyses were performed for the outcomes of flare and persistent flare.

Further, in order to evaluate the robustness of the findings in the primary analyses, an arm-based generalized linear mixed model (empirical Bayes) was used to evaluate the odds for flare and persistent flare for the tapering, withdrawal and continuation groups. All analyses were performed in Stata (version 16; StataCorp, College Station, TX, USA) or SAS Studio (version 9.4; SAS Institute, Cary, NC, USA).

Grading the certainty of the evidence

The Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach was used to assess the overall quality of the evidence for risk of bias, imprecision, inconsistency, indirectness and publication bias [27]. The GRADE ratings of very low, low, moderate or high quality evidence was applied [28].

Results

Study selection

The systematic literature search identified 14 861 references. Fig.  1 provides a detailed overview of study selection. From 163 articles assessed in full text, 23 studies were found eligible and included in the qualitative synthesis; however, only 22 studies were included in the meta-analyses, as one study did not provide sufficient data on the control group [29].

Fig. 1

Flow diagram of study selection

Study characteristics and risk of bias in individual trials

The study characteristics of the 23 included trials published between 2012 and 2021 are presented in Table  1. The study population based on data from the 22 trials included in the meta-analyses comprised 4082 patients with RA from 17 trials and 831 patients with axSpA from 5 trials. The axSpA population were AS in three studies [31, 33, 34], axSpA including AS in one study [30] and non-radiographic axSpA in one study [32]. No studies were identified for patients with PsA or a mixed population of IA. Of the included trials, 64% were funded by a drug manufacturer [29, 30, 32, 33, 35–37, 39, 40, 43, 44, 46–49], with no difference between RA or axSpA [12/18 trials (67%) vs 3/5 trials (60%), respectively]. However, significantly more withdrawal studies [10/11 trials (91%)] were funded by a manufacturer compared with tapering trials [8/16 trials (50%)].

Table 1

Study characteristics and risk of bias assessment of trials eligible for qualitative synthesis

First author, acronym, year	Disease duration, years	Inclusion criteria before tapering/withdrawal	Strategy	Randomised, n	Primary endpoint assessment	Definition of flare	Risk of biasa
axSpA trials
Landewé, C-OPTIMISE, 2020 [30]	TG: 2.0 WG: 2.1 CG: 2.5	ASDAS <1.3 at week 32 or 36 and week 48 in period 1	TG: CZP 200 mg every 4 weeks WG: CZP stop CG: CZP 200 mg every 2 weeks	TG: 105 WG: 104 CG: 104	48 weeks	ASDAS ≥2.1 at two consecutive visits or ASDAS >3.5 at any visit
Gratacos, REDES-TNF, 2019 [31]	TG: 9.3 CG: 10.4	BASDAI ≤2, no arthritis or enthesitis and no CRP ≥URL ≥6 months	TG: ADA 40 mg every 3 weeks, ETN 50 mg every 10 days, GOL 50 mg every 6 weeks or IFX 3 mg/kg every 8 weeks CG: standard dose TNFi continued	TG: 61 CG: 62	52 weeks	ASDAS ≥2.1 or BASDAI and physician GA ≥4 and patient GA ≥4, nocturnal axial pain ≥4 or increased CRP/ESR
Landewé, ABILITY-3, 2018 [32]	WG: 1.8 CG: 1.9	Sustained remission (ASDAS score <1.3) at weeks 16, 20, 24 and 28 in period 1	WG: ADA stop CG: ADA 40 mg every 2 weeks	WG: 153 CG: 152	40 weeks	ASDAS ≥2.1 on two consecutive visits
Yates, ANSWERS, 2015 [33]	TG: NA CG: NA	Responders (50% reduction or ≥2 units decrease in BASDAI and spinal pain. Duration not specified	TG: ETN 25 mg/week CG: ETN 50 mg/week	TG: 23 CG: 24	26 weeks	ΔBASDAI ≥2 or 50% increase to baseline on two or more visits and Δspinal pain ≥2 and reinstatement of therapy needed
Cantini, 2013 [34]	TG: 13 CG: 12	BASDAI <4, no arthritis, dactylitis, tenosynovitis or uveitis and normal ESR/CRP. Duration not specified	TG: ETN 50 mg every 2 weeks CG: ETN 50 mg/week	TG: 22 CG: 21	TG: mean 22 months (s.d. 1) CG: mean 21 months (s.d. 1.6)	BASDAI >4 or extra-articular manifestations
RA trials
Curtis, SEAM-RA, 2020 [35]	WG: 9.7 CG: 10.3	SDAI ≤3.3 through 24 weeks	WG: ETN stop CG: ETN 50 mg/week	WG: 101 CG: 51	48 weeks	ΔSDAI 3.4–11.0 on two visits ≥2 weeks apart or SDAI 3.4–11.0 at any time on three or more separate visits or any SDAI >11
Sanmarti, TO-SPACE, 2019 [36]	TG: 6.4 CG: 6.5	DAS28-ESR <2.6 between week 20 and 24 in period 1	TG: TCZ 162 mg every 2 weeks CG: TCZ 162 mg/week	TG: 90 CG: 89	24 weeks	DAS28-ESR score ≥2.6
Takeuchi, RA-BEYOND, 2019 [37]	TG: 9.3 CG: 9.5	≥3 months: CDAI ≤10 for RA-BEAM, -BUILD, -BEACON or CDAI ≤2.8 for RA-BEGIN	TG: BCN 2 mg/day CG: BCN 4 mg/day	TG: 278 CG: 281	48 weeks	Failure to maintain CDAI LDA or remission
Verhoef, REDO, 2019 [25]	TG: 14.0 CG: 16.9	≥6 months LDA after last RTX (DAS28-CRP <2.9 or LDA judged by rheumatologist) and DAS28-CRP ≤3.5 at baseline	TG: RTX 500 mg once or RTX 200 mg once CG: RTX 1000 mg once	TG: 113 CG: 29	26 weeks	ΔDAS28-CRP >1.2 from baseline or ΔDAS28-CRP >0.6 from baseline and DAS28-CRP ≥2.9 at month 6
l’Ami, 2018 [38]	TG: 11 CG: 11	ADA serum trough concentration >8 μg/mL after at least 28 weeks of treatment	TG: ADA 40 mg every 3 weeks CG: ADA 40 mg every 2 weeks	TG: 27 CG: 28	28 weeks	DAS28-ESR ≥0.6
Ibrahim, OPTTIRA, 2017 [26]	TG: 11.6 CG: 10.6	Sustained DAS28-ESR <3.2 without an increase of >0.6 during the last 3 months	TG: ADA/ETN tapered by 33% or ADA/ETN tapered by 66% CG: ADA 40 mg every 2 weeks or ETN 50 mg/week	TG: 47 CG: 50	26 weeks	ΔDAS28-ESR ≥0.6 with DAS28 > 3.2 and increased swollen joint on two visits ≥1 week apart or ΔDAS28-ESR ≥1.2 with DAS28 > 3.2
Pavelka, 2017 [39]	WG: 8.3 CG: 8.0	LDA (DAS28-ESR <3.2) at week 24 in period 1	WG: ETN stop CG: ETN 50 mg/week	WG: 177 CG: 169	28 weeks	DAS28-ESR ≥3.2 and ΔDAS28-ESR ≥0.6 from the week 24
Weinblatt, C-EARLY, 2017 [40]	TG: 2.6 WG: 2.5 CG: 2.9	Sustained DAS28-ESR ≤3.2 at weeks 40 and 52 in period 1	TG: CZP 200 mg every 4 weeks WG: CZP stop CG: CZP 200 mg every 2 weeks	TG: 127 WG: 82 CG: 84	52 weeks	Self-reported flare and at 2 visits 2 weeks apart ΔDAS28-ESR ≥0.6 from period 2 baseline and DAS28-ESR >3.2 and investigator-judged RA activity
Fautrel, STRASS, 2016 [41]	TG: 8.3 CG: 11.0	DAS28-ESR ≤2.6 for 6 months and no progression on X-rays during the last year	TG: stepwise increased ADA/ETN injection interval every 3 monthsb CG: ADA 40 mg every 2 weeks or ETN 50 mg/week	TG: 64 CG: 74	78 weeks	DAS28-ESR >2.6 AND ΔDAS28-ESR >0.6 since the last visit
Ghiti Moghadam, POET, 2016 [42]	WG: 12.0 CG: 11.1	In last 6 months: DAS28-CRP <3.2 or LDA judged by rheumatologist and DAS28-CRP <3.2 at baseline and one or more CRP <10 mg/L	WG: stop TNFi CG: standard dose TNFi continued	WG: 531 CG: 286	52 weeks	DAS28 ≥3.2 and ΔDAS28 ≥0.6 compared with the baseline
Van Vollenhoven, DOSERA, 2016 [43]	TG: 16.6 WG: 11.5 CG: 12.3	DAS28-ESR ≤3.2 ≥11 months prior to baseline	TG: ETN 25 mg/week WG: ETN stop CG: ETN 50 mg/week	TG: 27 WG: 23 CG: 23	48 weeks	DAS28-ESR >5.1, DAS28-ESR >3.2 and ΔDAS28 ≥1.2 from baseline, DAS28-ESR >3.2 and ΔDAS28 ≥0.6 from baseline on two visits 1–3 weeks apart and disease flare judged by investigator or patient
Yamanaka, ENCOURAGE, 2016 [44]	WG: 2.4 CG: 1.9	DAS28 <2.6 at both month 6 and 12 in period 1	WG: ETN stop CG: ETN 50 mg/week	WG: 50 CG: 49	52 weeks	DAS28 ≥3.2
Raffeiner, 2015 [29]	TG: 14.3 CG: 13.4	DAS28-ESR <2.6 for at least 12 months	TG: ETN 25 mg/week CG: ETN 25 mg biweekly	TG: 159 CG: 164	Mean 3.6 years (s.d. 1.5)	DAS28-ESR >2.6
Van Herwaarden, DRESS, 2015 [45]	TG: 10 CG: 10	DAS28-CRP <3.2 or LDA judged by rheumatologist at two visits ≥3 months apart	TG: stepwise increased ADA/ETN injection interval every 3 monthsc CG: ADA 40 mg every 2 weeks or ETN 50 mg/week	TG: 121 CG: 59	78 weeks	ΔDAS28-CRP >1.2 from baseline or ΔDAS28-CRP >0.6 from baseline and current DAS28 ≥3.2
Westhovens, AGREE, 2015 [46]	TG: 2.4 CG: 2.4	DAS28-ESR <2.6 on day 701 (i.e. year 2 of phase 1)	TG: ABA 5 mg/kg every 4 weeks CG: ABA 10 mg/kg every 4 weeks	TG: 50 CG: 58	52 weeks	Additional DMARD required or two or more courses of high-dose steroids or require open-label ABA 10 mg/kg or DAS28-CRP ≥3.2 at two consecutive visits
Smolen, OPTIMA, 2014 [47]	WG: 0.8 CG: 0.8	DAS28-CRP ≤3.2 at weeks 22 and 26 in period 1	WG: ADA stop CG: ADA 40 mg every 2 weeks	WG: 102 CG: 105	52 weeks	Not defined
Smolen, PRESERVE, 2013 [48]	TG: 6.4 WG: 6.8 CG: 7.3	DAS28-ESR ≤3.2 from weeks 12–36 in period 1 and DAS28-ESR ≤3.2 at week 36	TG: ETN 25 mg/week WG: ETN stop CG: ETN 50 mg/week	TG: 202 WG: 200 CG: 202	52 weeks	DAS28-ESR >3.2 and ΔDAS28 ≥0.6 or discontinuation due to poor efficacy, protocol violation or other
Chatzidionysiou, ADMIRE, 2012 [49]	WG: 10.4 CG: 7.6	Sustained remission (DAS28 <2.6) ≥3 months	WG: ADA stop CG: ADA 40 mg every 2 weeks	WG: 16 CG: 17	28 weeks	DAS28 ≥2.6 or ΔDAS28 >1.2 from baseline

a

RoB 2.0 consists of the following assessments represented by a coloured bullet: risk of bias arising from the randomization process, risk of bias due to deviations from the intended interventions, missing outcome data, risk of bias in measurement of the outcome, risk of bias in selection of the reported result and overall risk of bias. The bullet colour represents the following risk of bias assessment: green, low risk of bias; yellow, unclear risk of bias; red, high risk of bias.

b

ETA 50 mg every 10 days (step 1), 14 days (step 2), 3 weeks (step 3) or stopped (step 4); ADA 40 mg every 21 days (step 1), 28 days (step 2), 6 weeks (step 3) or stopped (step 4).

c

ADA: 40 mg every 21 days, 40 mg every 28 days and stop; ETN: 50 mg every 10 days, 50 mg every 14 days and stop.

TG: tapering group; WG: withdrawal group; CG: continuation group; CZP: certolizumab pegol; URL: upper reference limit; ADA: adalimumab; ETN: etanercept; GOL: golimumab; IFX: infliximab; GA: global assessment; NA: not assessed; SDAI: Simplified Disease Activity Index; TCZ: tocilizumab; CDAI: Clinical Disease Activity Index; BCN: baricitinib; RTX: rituximab; ABA: abatacept.

The mean age across trials was 49 years, with a younger axSpA population (39 years) compared with the RA population (52 years). As expected, a higher proportion of females were observed in the RA population (71%) compared with the axSpA population (22%). Disease duration was between 10 months [47] and 16.9 years [25].

The results of the risk of bias assessment of the 23 included studies according to the RoB 2.0 are presented in Table  1. Only one study was evaluated that corresponded to a low risk of bias in all items [24].

Synthesis of results

Risk for flare

Fifteen studies were included in a meta-analysis of the risk for flare after tapering b-/tsDMARDs compared with continuation of the standard dose. Eight trials assessed etanercept [26, 31, 33, 34, 41, 43, 45, 48], five adalimumab [26, 31, 38, 41, 45], two certolizumab pegol [30, 40] and one each for abatacept [46], baricitinib [37], golimumab [31], infliximab [31], rituximab [25] and tocilizumab [36]. Across the 15 trials, the observed flare rate was 32.7% (431/1319) for tapering and 22.2% (254/1146) for continuation. As presented in Fig.  2, the meta-analysis of these data demonstrated an increased risk of flares [RR 1.45 (95% CI 1.19, 1.77)], corresponding to an absolute number needed to treat in order to harm (NNH) of 10 patients.

Fig. 2

Random effects meta-analysis of flare risk after either tapering or withdrawal of b-/tsDMARDs compared with continuation

In a meta-analysis assessing flare risk after withdrawal compared with continuation of the standard dose, only data on TNFis were available. Among 11 included studies, 6 evaluated etanercept [35, 39, 42–44, 48], 4 adalimumab [32, 42, 47, 49], 3 certolizumab pegol [30, 40, 42] and one study each for golimumab [42] and infliximab [42]. The observed flare rate was 55.3% (830/1500) for withdrawal and 24.1% (293/1217) for continuation. In the meta-analysis, a highly increased flare risk (Fig.  2) was observed [RR 2.28 (95% CI 1.78, 2.93)], with an NNH of three patients.

As presented in Supplementary Table S1 (available at Rheumatology online), a network meta-analysis confirmed the highly increased odds for flare when b-/tsDMARDs were withdrawn compared with tapered [OR 5.62 (95% CI 3.44, 9.17)].

Risk for persistent flare

Eight trials assessing a tapering strategy to continuation of the standard dose had data on persistent flare and were included in the meta-analysis. Four trials assessed etanercept [41, 43, 45, 48], two adalimumab [41, 45], two certolizumab pegol [30, 40] and one trial each for baricitinib [37] and rituximab [25]. The observed persistent flare rate was 4.8% (50/1035) for tapering and 2.9% (25/854) for continuation. However, the meta-analysis only indicated potentially increased odds for persistent flare [POR 1.56 (95% CI 0.97, 2.52)] (Fig.  3), with an NNH of 64 patients.

Fig. 3

Random effects meta-analysis of persistent flare odds after either b-/tsDMARDs tapering or withdrawal compared with continuation

Seven trials evaluating a withdrawal strategy of TNFi compared with continuation of the standard dose had data on persistent flare. Four trials evaluated etanercept [35–37, 46], two certolizumab pegol [30, 40] and one adalimumab [32]. The observed persistent flare rate was 15.3% (108/707) for withdrawal and 4.2% (28/664) for continuation. The meta-analysis demonstrated a significant increase in odds for persistent flare (Fig.  3) [POR 3.41 (95% CI 1.91, 6.09)], with an NNH corresponding to 11 patients.

The network meta-analysis confirmed the increased odds for persistent flare when comparing withdrawal of b-/tsDMARDs to tapering [OR 3.16 (95% CI 1.49, 6.67)] (Supplementary Table S1, available at Rheumatology online).

No statistically significant difference in odds for SAEs, serious infections or death was observed between the tapering group and the continuation group or between the withdrawal group and the continuation group (Supplementary Figs S1–S3, available at Rheumatology online).

Sensitivity analyses

Stratified meta-analyses on the risk for flare showed no significant difference between patients with RA or axSpA tapering or withdrawing b-/tsDMARDs compared with continuation of the standard dose (Supplementary Fig. S4, available at Rheumatology online). However, when comparing withdrawal of TNFis with continuation, the risk for persistent flare in patients with RA appeared to be higher than for patients with axSpA (Supplementary Fig. S5, available at Rheumatology online).

When stratifying for b-/tsDMARD therapy, a subgroup meta-analysis demonstrated an increased flare risk when baricitinib, tocilizumab or a combined group of TNFis (i.e. studies with two or more different TNFis) were tapered compared with continuation of the standard dose (Supplementary Fig. S6, available at Rheumatology online) but not when tapering abatacept, adalimumab, certolizumab pegol, etanercept or rituximab. When assessing withdrawal, an increased risk for flare was demonstrated for adalimumab, etanercept and a combined group of TNFi, but not for certolizumab pegol (Supplementary Fig. S6, available at Rheumatology online). However, as data were sparse for the individual drugs for both tapering and withdrawal, these subanalyses are limited by uncertainties.

When the risk for flare in the tapering group compared with the continuation group was stratified by dose reduction strategy, the flare risk appeared to be highest for the disease activity–guided dose reduction strategy, followed by the one-step fixed 50–65% dose reduction strategy, and lowest for the one-step fixed 25% dose reduction strategy (Supplementary Fig. S7, available at Rheumatology online).

Stratifying for time since arthritis diagnosis, a trend towards a higher flare risk for patients with established disease (i.e. diagnosis ≥3 years) compared with those recently diagnosed (i.e. <3 years) was observed when b-/tsDMARDs was tapered compared with continued (Supplementary Fig. S8, available at Rheumatology online). No significant difference in flare risk stratified by time since arthritis diagnosis was found when comparing TNFi withdrawal to continuation. Moreover, subgroup meta-analyses found no significant difference in flare risk when tapering or withdrawing b-/tsDMARDs compared with continuation of the standard dose were stratified by length of prior remission/LDA (i.e. <26 vs ≥26 weeks) before the start of the intervention period (Supplementary Fig. S9, available at Rheumatology online). However, a significantly increased flare risk when withdrawing TNFis compared with continuation was observed in studies with a longer follow-up period (i.e. a longer intervention period) compared with studies with a shorter follow-up period (Supplementary Fig. S10, available at Rheumatology online). No significant difference in flare risk was observed between the tapering group and the continuation group when stratifying for the length of the intervention period. Lastly, a list of potential prognostic factors is presented in Supplementary Table S2, available at Rheumatology online.

Certainty of the evidence

Quality assessment of the evidence was performed in accordance with the GRADE approach [27] and presented in Table  2. Visual assessment in a funnel plot did not reveal any publication bias for the primary outcome of flare (Supplementary Fig. S11, available at Rheumatology online), nor did an Eggers test for small-study effects. A large difference in studies funded by the manufacturer was observed (withdrawal trials, 91%; tapering trials, 50%); however, a stratified meta-analysis did not reveal any difference in flare risk between studies that were funded by the manufacturer or not (Supplementary Fig. S12, available at Rheumatology online).

Table 2

Summary of evidence synthesis: tapering/withdrawal group vs continuation group

Summary of findings						Quality assessment
Outcome (n studies)	Reduction group, n/N (%)	Continuation group, n/N (%)	Relative measure (95% CI)	Absolute measure (95% CI)	NNH	Overall risk of biasa	Inconsistencyb, I2 (%)	Indirectness	Imprecisionc	Publication biasd	GRADE quality
Tapering vs continuation group
Flare (15 studies)	431/1319 (32.7)	254/1146 (22.2)	RR 1.45 (1.19, 1.77)	99.2/1000 (46.8, 151.7)	10	Serious	Not serious, 42.5	Not serious	Not serious	Undetected	Moderate
Persistent flare (8 studies)	50/1035 (4.8)	25/854 (2.9)	POR 1.56 (0.97, 2.52)	15.6/ 1000 (−1.2, 32.3)	64	Serious	Not serious, 0.0	Not serious	Serious	Undetected	Low
Serious infection (11 studies)	12/1121 (1.1)	16/945 (1.7)	POR 0.51 (0.24, 1.10)	−8.3/ 1000 (−17.7, 1.1)	−121	Not serious	Not serious, 0.0	Not serious	Not serious	Undetected	High
SAE (14 studies)	95/1334 (7.1)	80/1161 (6.9)	POR 0.91 (0.61, 1.35)	−6.0/1000 (−30.6, 18.6)	−166	Not serious	Not serious, 23.2	Not serious	Serious	Undetected	Moderate
Death (9 studies)	1/862 (0.1)	2/674 (0.3)	POR 0.58 (0.09, 3.84)	−1.3/1000 (−5.6, 3.1)	−793	Not serious	Not serious, 0.0	Not serious	Very serious	Undetected	Low
Flare in axSpA (4 studies)	48/210 (22.9)	32/209 (15.3)	RR 1.49 (0.99, 2.24)	75.4/1000 (−1.15, 152.0)	13	Not serious	Not serious, 0.0	Not serious	Serious	Undetected	Moderate
Flare in RA (11 trials)	383/1109 (34.5)	222/937 (23.7)	RR 1.42 (1.11, 1.81)	98.6/1000 (28.8, 168.4)	10	Serious	Not serious, 59.1	Not serious	Not serious	Undetected	Moderate
Withdrawal vs continuation group
Flare (11 studies)	830/1500 (55.3)	293/1217 (24.1)	RR 2.28 (1.78, 2.93)	309.4/1000 (216.6, 402.4)	3	Serious	Serious, 78.1	Not serious	Not serious	Undetected	Low
Persistent flare (7 studies)	108/707 (15.3)	28/664 (4.2)	POR 3.41 (1.91, 6.09)	88.0/ 1000 (46.3, 129.7)	11	Serious	Not serious, 48.6	Not serious	Not serious	Undetected	Moderate
Serious infection (8 studies)	25/1282 (2)	14/1035 (1.4)	POR 1.43 (0.75, 2.71)	5.9/1000 (−4.7, 16.5)	170	Not serious	Not serious, 0.0	Not serious	Serious	Undetected	Moderate
SAE (10 studies)	90/1486 (6.1)	47/1192 (3.9)	POR 1.49 (0.79, 2.83)	18.1/1000 (−10.9, 47.1)	55	Not serious	Not serious, 62.3	Not serious	Serious	Undetected	Moderate
Death (10 studies)	0/1486 (0)	3/1192 (0.3)	POR 0.23 (0.04, 1.43)	−2.3/1000 (−5.2, 0.6)	−431	Not serious	Not serious, 0.0	Not serious	Very serious	Undetected	Low
Flare in axSpA (2 studies)	164/257 (63.8)	62/256 (24.2)	RR 2.91 (1.09, 7.79)	462.5/ 1000 (36.6, 888.3)	2	Not serious	Serious, 92.8	Not serious	Not serious	Undetected	Moderate
Flare in RA (9 studies)	666/1243 (53.6)	231/961 (24.0)	RR 2.16 (1.69, 2.75)	277.9/1000 (189.7, 366.1)	4	Serious	Not serious, 69.6	Not serious	Not serious	Undetected	Moderate

a

Evaluated using subgroup meta-analysis stratified for the overall risk of bias assessment for the primary outcome flare and the secondary outcome persistent flare. The overall risk of bias for adverse events was judged not to be serious based on expert opinion.

b

I² test was evaluated as <25% = not serious and >75% = serious. I² test result between 25 and 75% was evaluated as serious if the relative estimate from the fixed effect meta-analysis was within the limits of the 95% CI from the random effects meta-analysis.

c

Evaluated as serious if the 95% CI of the relative measure was inconclusive.

d

Evaluated for the primary outcome flare by funnel plots, an Egger’s test and subgroup meta-analyses stratified by funding by the manufacturer. Publication bias for adverse events was judged not to be serious based on expert opinion.

Withdrawing bDMARDs compared with continuation of the standard dose stratified by the overall risk of bias showed a significant difference, as studies with a high overall risk of bias had a higher flare risk and persistent flare risk compared with studies with an unclear overall risk of bias (Supplementary Figs S13 and S14, available at Rheumatology online). The meta-analyses did not reveal any significant group differences for tapering compared with continuation; however, a trend towards a higher flare risk for studies with less overall risk of bias and a trend towards a higher persistent flare risk for studies with higher overall bias was observed.

Discussion

To our knowledge, this SLR is the first to evaluate tapering and withdrawal of various b-/tsDMARDs in patients with RA and axSpA. An increased flare risk for patients who taper or withdraw b-/tsDMARDs compared with continuation of the standard dose was observed; however, the odds for persistent flare were only significantly increased for withdrawal compared with continuation. Moreover, a network meta-analysis demonstrated a highly significant increased odds for flare and persistent flare when withdrawal of b-/tsDMARDs was compared with tapering.

The findings in this SLR support current international treatment guidelines in RA and axSpA, as tapering, but not withdrawal, is judged to be feasible and safe [4, 5]. Moreover, the findings are in line with previous evidence, as a recent Cochrane SLR demonstrated that withdrawal of TNFi in patients with RA may be inferior to continuation, but tapering was comparable [9]. Similarly, Henaux et al. [10] found an increased risk among patients with RA for loss of LDA when withdrawing bDMARDs compared with continuation of the standard dose but no increased risk when tapering. A recent review by Vasconcelos et al. [11] supports these findings, as no significant difference in LDA in patients with RA was observed between the tapering group and the continuation group. Furthermore, Schlager et al. [12] reported very high odds for flare after withdrawal of bDMARDs compared with continuation in patients with RA. The effect of tapering or withdrawal of b-/tsDMARDs on disease activity in patients with axSpA have not yet been evaluated in a meta-analysis. However, the observed flare rate in this SLR of 63.8% for patients with axSpA withdrawing TNFi lies within the range of previously reported flare rates of 11–53% [15] and 76–100% [13]. Three studies that were excluded in this review after full-text screening have previously been included in other reviews; however, these studies had only a small proportion of patients in treatment with bDMARDs [50, 51] or the follow-up period was very short (<24 weeks) [52].

This SLR demonstrated no significant difference in odds for SAEs, serious infections or death when tapering or withdrawing b-/tsDMARDs compared with continuation of the standard dose. Similarly, a recent meta-analysis by Vinson et al. [53] found no significant difference in the risk for serious infections, SAEs, death, malignancies or adverse events among patients with RA or axSpA who tapered or continued b-/tsDMARDs.

There are limitations to this work, as the majority of studies had data on patients with RA (17 studies), with only 5 studies on axSpA and none on PsA. However, no significant difference in flare risk was observed between patients with RA or axSpA tapering or withdrawing b-/tsDMARDs compared with continuation of the standard dose. No conclusion can be drawn from this work on the flare risk among patients with PsA, due to the absence of available studies; future research is needed. Another important limitation to discuss is the heterogeneous flare definition in the included trials, as validated flare criteria were first published in 2013 for RA [7] and 2018 for axSpA [8]. However, the two independent assessors (L.U. and S.K.) judged the flare definitions to be adequate and reasonable for measuring flare, thereby allowing combination of trial data.

Important differences in tapering strategies (e.g. one-step fixed-dose reduction vs disease activity–guided dose reduction) was observed among the included studies. However, these differences were taken into consideration in a subgroup meta-analysis where a tendency towards a higher flare risk (as expected) was observed for the disease activity–guided dose reduction strategy (step-wise reduction of b-/tsDMARDs until flare or complete withdrawal) compared with a one-step fixed-dose reduction strategy.

Data on b-/tsDMARDs with different modes of action were combined in the meta-analyses, as data on the individual drugs are sparse; however, subgroup meta-analyses were performed to explore potential differences in flare risk. The subgroup analyses revealed that some drugs seem to have a greater flare risk than others. However, conclusions must be made with caution, as data are sparse, especially for abatacept, baricitinib, rituximab and tocilizumab, where only one study exists.

Another limitation to discuss is the heterogeneity in study designs: newly diagnosed or established disease, the duration of remission/LDA before start of the intervention period and the duration of follow-up. Subgroup meta-analyses were performed to explore these differences and revealed a trend towards a higher flare risk in patients with established disease when tapering b-/tsDMARDs compared with continuation and a higher flare risk in studies with a longer follow-up period when TNFi was withdrawn compared with continued. Thus the observed risk differences are important to keep in mind when discussing dose reduction in clinical practise.

The overall certainty of evidence was judged to be low for the following outcomes: flare when withdrawing, persistent flare when tapering and death for both tapering and withdrawal. Thus this limits the confidence of these findings and highlights the need for future research.

Conclusion

To our knowledge, this meta-analysis is the first to evaluate the risk of flare among patients with RA and axSpA tapering or withdrawing various b-/tsDMARDs compared with continuation of the standard dose. A significant and noteworthy high risk for flare and, moreover, for persistent flare was observed when withdrawing TNFi. Furthermore, an increased risk for flare but not for persistent flare was demonstrated when b-/tsDMARDs were tapered. Thus for clinical practise, tapering should be considered the more favourable approach (compared with withdrawal) when dose reduction of b-/tsDMARDs is considered feasible.

Acknowledgements

This SLR was based on research using data from AbbVie that was made available through Vivli. Vivli has not contributed to or approved and is not in any way responsible for the content of this publication. L.U. drafted the manuscript. L.U., S.K., L.D., R.C., A.S. and E.M.H. contributed to development of the review design. L.U. and C.S. developed the search strategy and C.S. performed the database searches. R.C. provided specialist statistical knowledge and drafted the data analysis plan. W.K.H.D., C.L. and S.S.A. performed the phase 1 screening. S.K. and L.U. performed phase 2 and 3 screening, data extraction and risk of bias assessment. L.U. and R.C. conducted the quality assessment in accordance with GRADE. All authors contributed to refinement of this systematic review and approved the final manuscript.

Funding: No specific funding was received from any bodies in the public, commercial or not-for profit sectors to carry out the work described in this article. The Parker Institute is supported by a core grant from the Oak Foundation (OCAY-18-774-OFIL). The Oak Foundation had no role in this work.

Disclosure statement: L.D. has received grants from Bristol Myers Squibb and speakers bureau fees from Galderma, Eli Lilly and Janssen. E.M.H. has received speakers bureau/consulting fees from AbbVie, Sanofi, Sobi and SynACT Pharma and research funding to Aarhus University Hospital from Novo Nordisk Foundation, Danish Rheumatism Association, Danish Regions Medicine Grants, Roche and Novartis. PCT has received grants from Celgene and Galapagos and consulting fees from AbbVie, Biogen, Eli Lilly, Fresenius, Galapagos, Gilead Sciences, GlaxoSmithKline, Janssen, Nordic Pharma, Pfizer, Roche and Sanofi. The remaining authors have nothing to disclose.

Data availability statement

This review contains data that are available in the included published papers, however, access to (not publicly available) study protocols and statistical analysis plans were obtained from the authors.

Supplementary data

Supplementary data are available at Rheumatology online.