A systematic review and network meta-analysis of the safety of early interventional treatments in rheumatoid arthritis

Abstract Objectives To evaluate the safety of treatment strategies in patients with early RA. Methods Systematic searches of MEDLINE, EMBASE and PubMed were conducted up to September 2020. Double-blind randomized controlled trials (RCTs) of licensed treatments conducted on completely naïve or MTX-naïve RA patients were included. Long-term extension studies, post-hoc and pooled analyses and RCTs with no comparator arm were excluded. Serious adverse events, serious infections and non-serious adverse events were extracted from all RCTs, and event rates in intervention and comparator arms were compared using meta-analysis and network meta-analysis (NMA). Results From an initial search of 3423 studies, 20 were included, involving 9202 patients. From the meta-analysis, the pooled incidence rates per 1000 patient-years for serious adverse events were 69.8 (95% CI: 64.9, 74.8), serious infections 18.9 (95% CI: 16.2, 21.6) and non-serious adverse events 1048.2 (95% CI: 1027.5, 1068.9). NMA showed that serious adverse event rates were higher with biologic monotherapy than with MTX monotherapy, rate ratio 1.39 (95% CI: 1.12, 1.73). Biologic monotherapy rates were higher than those for MTX and steroid therapy, rate ratio 3.22 (95% CI: 1.47, 7.07). Biologic monotherapy had a higher adverse event rate than biologic combination therapy, rate ratio 1.26 (95% CI: 1.02, 1.54). NMA showed no significant difference between strategies with respect to serious infections and non-serious adverse events rates. Conclusion The study revealed the different risk profiles for various early RA treatment strategies. Observed differences were overall small, and in contrast to the findings of established RA studies, steroid-based regimens did not emerge as more harmful.


, 21.6) and n
n-serious adverse events 1048.2 (95% CI: 1027.5, 1068.9).NMA showed that serious adverse event rates were higher with biologic monotherapy than with MTX monotherapy, rate ratio 1.39 (95% CI: 1.12, 1.73).Biologic monotherapy rates were higher than those for MTX and steroid therapy, rate rati 3.22 (95% CI: 1.47, 7.07).Biologic monotherapy had a higher adverse event rate than biologic combination therapy, rate ratio 1.26 (95% CI: 1.02, 1.54).NMA showed no significant difference between strategies with respect to serious infections and n n-serious adverse events rates.Conclusion.The study revealed the different risk profiles for various early RA treatment strategies.Observed differences were overall small, and in contrast to the findings of established RA studies, steroid-based regimens did not emerge as more harmful.

Introduction

RA has a prevalence of 0.5-1.0%[1].It is characterized by chronic joint inflam ation, synovial hyperplasia and systemic manifestations.Without adequate treatment, RA can lead to severe joint deformity and disability [2,3], impacting upon patients' quality of life and work ability [4,5].


Rheumatology key messages

. For the typical early RA patient, few differences exist between treatments strategies in erms of risk profile.

. Steroid-based treatment strategies did not emerge as more harmful in early RA compared to established RA.

The historic approach to treating newly diagnosed RA involved bed rest, analgesia, and (subsequently) DMARDS [6].The introduction of treat-to-target (T2T) strategies optimized outcomes for patients, with disease activity guiding adjustments in treatment with conventional synthetic DMARDs (csDMARDs), biologic DMARDs (bDMARDs), targeted synthetic DMARDs (tsDMARDs) and CSs [7].

Patients with RA have increased morbidity and mortality relative to that of the general population, and RA associates with multiple comorbidities, including diabetes, infections and cardiovascular diseases [8].Potential side effects must therefore be ta en into consideration when commencing treatments in patients with RA [9].

There is little evidence available that specifically considers the safety of treatments for early RA.Most evidence is derived from people with well-established disease and a substantial burden of comorbidities.In 2021, we now treat RA more aggressively, an

al outcomes are much improved
As a consequence, comparing patients withestablished RA from previous decades to contemporary patients diagnosed in recent years lacks face validity.Understanding treatment risks and benefits early in the course of the disease is important to inform choices about the initial treatment strategy.This will not only help patients and clinicians select the right treatment first time, but also provide an evidence base that is generalizable to the current decade.

This systematic review and meta-analysis investigates the safety outcomes of the various treatm nt strategies used to treat RA in the existing literature.Our objective was to compare the safety profiles of the early RA treatment strategies, including monotherapy, combination therapy and therapy with and without glucocorticoids.


Methods


Database and search strategy

A systematic literature search was performed for human studies using MEDLINE, EMBASE and PubMed databases.Disease search ter

included RA, inflamm
tory arthritis, early arthritis; drug search terms included MTX, SSZ, LEF, Adalimumab, Certolizumab pegol, Etanercept, Golimumab, Infliximab, Tofacitinib, Baricitinib, Upadacitinib, Rituximab, Abatacept, Tocilizumab, and Sarilumab.The full search strategy was published online in advance.The initial search was conducted by two investigators (M.A. and V.A.) with verification from a third reviewer (J.G.).

The primary search was undertaken in April 2020, and a final search was performed in September 2020 to identify new studies that could be incorporated in the review.The study was performed in accordance with the preferred reporting system for systematic reviews (PRISMA) [10], and registered with the international prospective regis er of systematic reviews (PROSPERO registration: CRD42020195766).


Eligibility criteria

Eligible studies were English language publications of randomized double-blind clinical trials in adult patients with RA commencing DMARD therapy with a controlled comparator arm.The initial protocol specified the following: (1) p

ients had to be
iagnosed with RA based on the 2010 EULAR/ ACR criteria or ACR 1987; (2) patients had to be treatment naı ¨ve (although any dose of steroids or NSAIDs were permitted); (3) RCTs had to use currently licensed drugs at licensed doses for RA; (4) RCTs had to include an active comparator arm.After the initial search, a protocol amendment was introduced to define treatment naı ¨ve as MTX naı ¨ve or referred to those patients who had received MTX for 4 weeks and had a disease duration of 2 ye

s.

Studies were
excluded if they were open label or single-blinded trials or conducted on patients with unidentified arthritis.Long-term extension, post-hoc and pooled analysis studies, conference abstracts, case reports, letters to the editor, review articles, case-control studies and cohort studies were also excluded.


Study selection

Records were managed in EndNote (EndNote X9, Australia).Duplicate articles were identified and removed.Two researchers (M.A. and V.A.) independently screened study titles and abstracts in Rayyan QCRI, Qatar, and eligible studies were selected for inclusion.Further data management was conducted in Microsoft Excel.Disagreements over study eligibility were resolved through discussion with the third reviewer (J G.).Included studies were exported to EndNote X9 for fulltext screening.


Data extraction

Data extraction was performed in Microsoft Excel, and included study source (author and publication date), study registration number, study characteristics (phase, randomization sequence, type of blinding, pooling of analysis, duration, country, and exclusion and inclusion criteria), patient demographics (sex,

ge, weight, height, B
I, smoking status), disease duration and activity score (DAS28) pre-and post-intervention, study intervention and comparator details (number of patients in each group, dosage and duration), concomitant steroid usage, type of treatment strategy, serious and nonserious adverse event counts, serious and non-serious infectious adverse event counts plus discontinuation due to adverse events and death.

Only studies with published adverse event rates were included.Supplementary Materials, available at Rheumatology online were reviewed.Trials registered on clinicaltrials.govwere hecked, for reported adverse events.Data were extracted by both primary reviewers (M.A., V.A.).Risk of bias and study quality were assessed at study and outcome level using the Cochrane Risk of Bias 2 (ROB2) Tool [11].


Statistical analysis

All statistical analyses were conducted using Stata 16 (StataCorp LLC, USA).Patient exposure years were calculated for intervention and comparator groups using sample size and study duration.The primary outcome of interest was serious adverse events rate, defined as any event associated with the use of i.v.antibioti s, hospital admission, discontinuation of the drug, or death.Secondary outcomes included serious infection and non-serious adverse events rates.The incidences of serious and non-serious adverse events, and serious infection were calculated for each study.

Network meta-analysis (NMA) was conducted to allow comparisons across treatment strategies, even where direct comparisons were sparse.NMA models included both completely drug-naı ¨ve, and MTX-naı ¨ve studies, and a network meta-regression was performed to explore the effect of naı ¨ve status on estimates.Surface under cu ulative ranking curve (SUCRA) [12] was estimated in order to indicate the likelihood of different treatment strategies being associated with more harm in regard to serious dverse events, serious infections and/or nonserious adverse events.

For sensitivity analyses, a separate meta-analysis was p

ta-analysis across all stud
es ( MTX-naı ¨ve and treatment-naı ¨ve trials) confirms the results of the NMA for direct comparisons.The pairwise meta-analysis for treatment strategies that had a common MTX arm showed a higher rate ratio of serious adverse events with biologic treatment strategies compared with MTX monotherapy, 1.40 (95% CI: 1.05, 1.87).No other significant differences were observed between other treatment strategies (Supplementary Fig. S2, available at Rheumatology online).When limiting the comparison to MTX-naı ¨ve trials (Supplementary Fig. S3, available at Rheumatology online), the above results were consistent.However, when limiting the comp

le 2.A SUCR
graph (Supplementary Fig. S7, available at Rheumatology online) and forest plots of the rate ratios for serious infections across all studies (Supplementary Fig. S8, available at Rheumatology online) and also limited to treatment status ((Supplementary Figs S9-S11, available at Rheumatology online) are presented in the Supplementary Data.


Non-serious adverse events

The non-serious adverse events network included 17 studies [16-23, 25-33 ] because 3 studies did not comment on the non-serious adverse events and had to be excluded [14,15,24].There were 4990 patients in the interventions arms and 3061 patients in the MTX monotherapy arms.The pooled incidence rate of non-serious infections events per 1000 patient-years across all studies was 1048.2 (95% CI: 1027.5, 1068.9).No significant differences between strategies were observed across studies (Supplementary Fig. S12, av ilable at Rheumatology online).Data from direct comparisons and the pairwise meta-analysis are demonstrated in Table 2.


FIG. 2 Network plot of comparison of treatment strategies

Network plots of comparison of treatment strategies with respect to serious adverse events, ser ous infections and adverse events (non-serious adverse events) in patients with RA.The size of the circles is proportional to the number of patients in each arm.The line widths are proportional to the number of studies in the comparison.bDMARD: biologic DMARD; tsDMARD: targeted synthetic DMARD.

The SUCRA graph (Supplementary Fig. S13, available at Rheumatology online) and pairwise meta-analysis results across all studies (Supplementary Fig. S14, available at Rheumatology online), when limited to MTXnaı ¨ve (Supplementary Fig. S15, available at Rheumatology online) and treatment-naı ¨ve patients (Supplementary Figs S16 and S17, available at Rheumatology online), are pr sented in the Supplementary Data.When limiting the sensitivity analysis to MTX-naı ¨ve trials , the risk of non-serious adverse events was lower with the combination of a bDMARD and MTX, compared with MTX monotherapy, 0.94 (95% CI: 0.89, 0.99).


Risk of bias

The risk of bias was assessed across different domains according to the Cochrane Collaboration's Risk of Bias tool [11].Of the included studies, 14/20 (70%) had an overall low risk of bias.Details on individual study bias assessment can be found in Supplementary Table S1, available at Rheumatology online.Trials with recent publication years were least likely to be at high risk of bias across the assessed domains.Older trials were more likely to be assessed as having high risk of bias and excluded from the analysis due to incomplete outcome reporting.

Visual inspection of funnel plots for serious adverse events, serious infections and non-serious adverse events showed no major asymmetry (Supplementary Figs S18-S20, available at Rheumatology online).Reporting bias, such as selective outcome reporting and publication bias, could not be ruled out due to large standard errors and low incidence rates (serious infections).


Discussion

This systematic review and NMA reports on the safety of treatment strategies using licensed drugs in early RA.Differences were seen in the rates of serious adverse events between MTX, bDMARD and steroid treatment.The most clinically relevant and striking finding is that the early CSs use and MTX comes out as a very safe FIG. 3 Network meta-analysis of the rate ratio of serious adverse events Network meta-analysis allows indirect comparisons of serious adverse events between treatment strategies.Treatment effects are described comparing the first ( left-hand) strategy with the second ( right-hand) strategy.The first seven strategies were compared w rect comparisons.bDMARD: biologic DMARD; tsDMARD: targeted synthetic DMARD.treatment strategy, and were superior in this regard to bDMARD strategies.

To our knowledge, this is the first systematic review and NMA to compare all treatment strategies in early RA.Previous systematic reviews reporting treatment safety in RA mainly focused on certain drug class and/ or included patients with long standing disease [3 ][35][36].

The NMA showed a statistically significant difference between bDMARD and MTX monotherapy with regards to serious adverse events, with a bDMARD being associated with more harm than MTX treatment.Variations within biologics classes, and substantial study heterogeneity due to differences in patient populations and adverse events reporting, limit the interpretation of the analyses.CIs were wide, with a low number of studies included.Ideally, we would have head-to-head comparisons for all key questions across the treatment strategies in RA, but this is not a feasible expectation.We therefore used NMA to provide insights into comparisons not assessed in clinical trials.

NMA also favoured MTX þ steroid and MTX þ bDMARD strategies with respect to serious adverse events rates.No differences were apparent when specifically looking into infections and non-serious adverse events.Clinical interpretation of the NMA results and in  Pooled incidence rate per 1 00 patient-years and rate ratios for serious adverse events, serious infections and non-serious adverse events across treatment strategies based on studies with direct comparisons.For strategies with zero events reported in one or both arms, a continuity correction of a fixed value (0.5) was added to each cell to create an event rate to allow odds ratio comparisons of the event rates between studies.Higher SUCRA values indicate a greater likelihood of a given treatment causing the least number of events, such that when the SUCRA value is 1, the treatment is the best, and when it is 0, it is the worst.I2 test reflects the the percentage of the variability in effect estimates that is due to heterogeneity.bDMARD: biologic DMARD; I2: I squa e statistics; NA: not applicable as only one study was available for that strategy; NMA: network meta-analysis; PMA: pairwise meta-analysis; SUCRA: Surface Under the Cumulative RAnking curve; tsDMARD: targeted synthetic DMARD.
- - - - - - - - - - - MTX þ S S Z - - - - - - - - - - - - MTX þ
particular the SUCRA rankings requires extreme caution.Study and population heterogeneity was substantial, and many individual studies were small and provided imprecise estimates of effect size.Crucially, the SUCRA rankings do not provide CIs for the rankings and must be interpreted alongside the ta ulated rate ratios.An a priori belief held by the investigators was that CSs strategies would have a less favourable safety profile, based upon the extensive data in established RA showing the negative impacts of CSs use [37,38].Our analyses presented here contradict our prior assumption.

Many national and international guidelines recommend against long term CSs use in RA but advocate the use of steroids in the induction of remission in early disease [39,40].The strategy of the use of early CS combined with MTX is much less costly than an early biologic strategy, and alongside our finding of the safety profile advantage that currently recommended treatment strategy for early RA is appropriate.

Results from the direct comparisons using pairwise meta-analysis of completely treatment naı ¨ve trials, did not demonstrate a significant difference in serious adverse events risk between MTX þ steroid and MTX þ bDMARD, rate ratio 0.47 (95% CI: 0.21, 1.02) (Supplementary Fig. S5, available at Rheumatology online).However, only one trial was included in this direct comparison [18].The NMA analysis, on the other hand, included evidence based on indirect comparisons from another study with a steroid arm [33] and multiple studies in the MTX þ bDMARD arm [14][15][16][17][18][19][20][21][22][23][24][25][26][27].The NMA uses information from these indirect comparisons and so provides a different estimate of the effects of MTX þ steroid and MTX þ bDMARD, rate ratio 0.39 (95%CI: 0.18, 0.83) (Fig. 3).It is important to note that although there is a difference in interpretation based on the P-va

es, the magnitude of the e
fect is consistent between both methods.

Over time, RCTs reporting guidelines have changed.In 2004, the clinical trials regulations act was set to legislate human trials approval, conduction, monitoring and reporting [41].In our analyses, steroid data come from historical studies, and that may affect how meaningful the results are.The differences observed in the included studies' regulation requirements and registration with clin cal trial numbers could be another explanation.Of the included trials, three trials had no clinical trials registration number, and two of these were published before 2004.Another five trials did not post their results on clinicaltrials.gov.The sort of magnitude we have seen could be explained by the study bias observed in our included studies.

The explanations for the observed differences may lie in a true effect, or be attributable to variations in study design, sampl size, patient selection, and follow-up.Patients enrolled in monotherapy arms are usually patients who failed MTX and MTX combination strategies; thus, these may be selecting patients with high risk.It is relevant to consider the face validity of the findings: the results suggesting a potential benefit of a biologic/MTX combination over biologic monotherapy feel at odds with clinical instinct.However, it is notable that this observation is not novel [42].

The only observed results for serious adverse events overall without differences to serious infections may in part be a reflection of limited study power and the fact that serious infections were rare.It is known that lack of statistical power may affect the interpretation of the analysis [43].Serious infections were uncommon and did not vary substantially across arms.Of the included trials involving SSZ and steroids strategies, only one commented on serious infections [18].The observation that we did observe differences for serious adverse events suggests that outcomes other than infection explain the differences.Our study did not explore non-infectious events, and this is an area for future research.

Non-serious adverse events were common across arms, and potentially offer greater power for detecting differences between strategies.In general, the nonserious events were not significantly different across any of the strategies.The lack of any difference indica ed by the non-serious adverse events analysis seems surprising, but may reflect the diagnostic heterogeneity of what is coded as non-serious adverse events in a clinical trial.An inference could be that non-serious adverse events are reported during the first year of drug marketing and not considered to be a useful discriminator for pharmacovigi ance [44].It has been published recently that nonserious adverse event collection and reporting in RCTs is inconsistent [45].Future research in non-serious adverse events may benefit from stratified analysis looking into non-serious adverse events of special interest.


Strengths and limitations

Treatment of RA has changed across the years, with patients now presenting to clinical practice in the early stages of the disease.A major strength of this work was that we included trials with early RA.An NMA approach was used to compare all treatment strategies in early RA.We limited the analysis to licensed doses of clinical importance, whereas previous p