Objectives To update the evidence for the safety of synthetic disease-modifying antirheumatic drugs (sDMARDs), glucocorticoids (GC) and biological DMARDs (bDMARDs) in patients with rheumatoid arthritis (RA) to inform the European League Against Rheumatism (EULAR) recommendations for the management of RA.
Methods Systematic literature review (SLR) of observational studies (including registries). Interventions were any bDMARD (anakinra, infliximab, etanercept, adalimumab, rituximab, abatacept, tocilizumab, golimumab or certolizumab pegol) or sDMARD (methotrexate, leflunomide, hydroxychloroquine, sulfasalazine, gold/auranofin, azathioprine, chlorambucil, chloroquine, cyclosporin, cyclophosphamide, mycophenolate, minocycline, penicillamine, tacrolimus or tofacitinib) and a comparator was required. Information on GCs was collected from the included studies. All safety outcomes were included.
Results Forty-nine observational studies addressing diverse safety outcomes of therapy with bDMARDs met eligibility criteria. Substantial heterogeneity precluded meta-analysis of any of the outcomes. Patients on tumour necrosis factor inhibitors (TNFi) compared to patients on conventional sDMARDs had a higher risk of serious infections (adjusted HR (aHR) 1.1–1.8), a higher risk of tuberculosis, and an increased risk of infection by herpes zoster cannot be excluded. Patients on TNFi did not have an increased risk for malignancies in general, lymphoma or non-melanoma skin cancer, but the risk of melanoma may be slightly increased (aHR 1.5). From the studies identified on conventional sDMARDs, no new safety signals were found.
Conclusions The findings from this SLR confirm the known safety pattern of sDMARDs and bDMARDs for the treatment of RA.
- Rheumatoid arthritis
- DMARDs (biologic)
- DMARDs (synthetic)
- outcomes research
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Many randomised controlled trials (RCTs) have been undertaken providing data on the short-term safety of disease-modifying antirheumatic drugs (DMARDs). Very recently, a new nomenclature that brings some order in the classification of DMARDs has been proposed. This nomenclature will be followed throughout the text.1 Briefly, DMARDs are classified into biological DMARDs (bDMARDs) and synthetic DMARDs (sDMARDs). bDMARDs include biological originator DMARDs (boDMARDs) and biosimilars (bsDMARDs). sDMARDs include conventional synthetic DMARDs (csDMARDs) comprising the traditional DMARDs (eg, methotrexate, sulfasalazine) and targeted synthetic DMARDs (tsDMARDS), such as the new Janus kinase-inhibitors.1
A milestone in the management of rheumatoid arthritis (RA) has been the development of registries for the long-term evaluation of the safety and effectiveness of both sDMARDs bDMARDs.2 These registries include groups of unselected patients treated according to routine care and followed-up irrespective of treatment continuation with a particular drug.2 Very recently, information from these registries has become available for treatments that are on the market for the last decade. For some newer targeted synthetic DMARDs, however, such as tofacitinib, only trial data are available to weigh safety. With so many developments in the treatment of RA, it is important to repetitively update the evidence on the safety of DMARDs.
The aim of this systematic literature review (SLR) was to update the evidence for the safety of synthetic and biological DMARDs in patients with RA in order to inform a European League Against Rheumatism (EULAR) Task Force charged with updating the EULAR recommendations for the management of RA with DMARDs.3 The results of this and two other SLRs4 ,5 were presented providing the Task Force with the current states of evidence. The safety of glucocorticoids (GCs) was not reviewed in detail, as there is a EULAR Task Force on GCs especially dedicated to this.6–8
Two separate searches were run, one for the safety of bDMARDs and another for the safety of sDMARDs. The research questions were structured according to a PICO format (Patients, Intervention, Comparator, Outcomes) and the eligible study types were defined.9 Participants were defined as adults at least 18 years old with a clinical diagnosis of RA. Studies containing patients with other diagnoses were eligible only if the results from patients with RA were presented separately. For the bDMARDs search, the intervention was defined as any bDMARD (all drugs (anakinra, infliximab, etanercept, adalimumab, rituximab, abatacept, tocilizumab, golimumab or certolizumab pegol), formulations, and duration) and the comparator as another bDMARD, an sDMARD, placebo or the general population. For the sDMARDs search, the intervention was defined as any sDMARD (including conventional sDMARDs (csDMARDs) (methotrexate, leflunomide, hydroxychloroquine, sulfasalazine, gold/auranofin, azathioprine, chlorambucil, chloroquine, cyclosporin, cyclophosphamide, mycophenolate, minocycline or penicillamine, tacrolimus) and tofacitinib) and the comparator as another sDMARD, combination therapy, GCs, bDMARDs, placebo or the general population. For outcomes, all safety outcomes were considered, namely infections (including serious infections, opportunistic infections including tuberculosis and herpes zoster), malignancies, deaths, cardiovascular disease (including congestive heart failure, coronary heart disease, stroke), change in lipid levels, impairment in renal function, elevation of liver enzymes, haematological abnormalities (including neutropenia), gastrointestinal effects, demyelinating disease, induction of autoimmune disease, and teratogenicity. The types of studies considered for inclusion were cohort studies/registries and study series with >30 cases, all published from 2009 onwards, since this SLR was performed in order to inform the EULAR Task Force responsible for the update of the 2010 EULAR recommendations for the treatment of RA. SLRs that were captured by the search were used to obtain references of original studies, which were included provided they fulfilled the eligibility criteria, but the SLRs were not in order to avoid duplication of information from the same study. Of note, all studies were required to include a comparator group. Descriptions of safety events in one group of patients without any comparator were not considered sufficient, as they do not allow for a proper interpretation of the impact of the intervention on the outcome. As mentioned, GCs were not included as interventions in this SLR. Nevertheless, due to their importance and their potential risks, especially when combined with DMARDs, information on GCs in the included studies was extracted.
The following computerised bibliographical databases were searched: Medline, Embase, The Cochrane Central Register of Controlled Trials (Central), until 8 April 2013, without language restrictions. Details on complete search strategies are provided in online supplementary text 1. In order to retrieve additional references, an additional search for systematic reviews was carried out in the Cochrane Database of Systematic Reviews and the Database of Abstracts of Reviews of Effects. References from included studies were screened in order to identify further studies to include.
Selection of studies, data extraction and assessment of risk of bias
Two reviewers independently assessed each title and abstract for suitability for inclusion in the review (for bDMARDs: SR and JLN, for sDMARDs: SR and CG-V), according to the predetermined selection criteria, followed by full-text article review where necessary. For included studies, they independently extracted data regarding inclusion and exclusion criteria, follow-up time, characteristics of study population, outcome definition, interventions and outcome measures (also for the subgroup with GCs combined, when separately reported) using a standardised data extraction form.
The two reviewers independently assessed the risk of bias of each included study using the ‘Hayden-tool’ which evaluates the following items: participation, attrition, prognostic factor measurement, outcome measurement, confounding and analysis.10 For all these steps, disagreements between reviewers were discussed, and a third reviewer (RL) was involved if necessary.
Of 10 559 articles (after de-duplication) that were screened, 49 articles were included (see flowchart in online supplementary figure S1). These focused on diverse outcomes: 22 on infections,11–32 12 on cancer,33–44 5 on mortality,45–49 3 on neurological events,50–52 2 on cardiovascular events,53 ,54 2 on gastrointestinal perforations,55 ,56 1 on venous thrombotic events,57 1 on elevation of liver enzymes,58 and 1 on new-onset psoriasis.59 From the 22 studies on infections, 9 addressed serious infections,11–19 5 skin infections20–24 (mainly infection by herpes zoster), 4 tuberculosis25–28 and 4 other types of infections.29–32 From the 12 studies on cancer, 5 addressed all types of cancer,33–37 1 cancer in patients with cancer history,38 2 lymphoma39 ,40 (plus 3 of the previous ones that also included lymphoma34–36), 2 non-melanoma skin cancer (NMSC)41 ,42 (plus one of the previous ones that also mentioned NMSC35), 1 melanoma43 and 1 pancreatic cancer.44
Due to multiple sources of heterogeneity (eg, different eligibility criteria, definitions of exposure, comparators, definitions for the outcome, methods of assessment and/or validation, etc), a meta-analysis could not be performed, and we present a summary of pertinent findings from the individual studies (tables 1 and 2, more details in online supplementary tables S1–S98). From the nine studies on the risk of serious infections, four of which at low risk of bias,11 ,12 ,17 ,19 all except for one (which included patients on anakinra19) included patients on the first three TNFi (infliximab, etanercept and adalimumab) only. Having csDMARDs as the comparator, adjusted HRs (aHR) for the risk of serious infections on TNFi between 1.1 and 1.8 were found, most of them reflecting a statistically significant risk. Five studies (3 at low risk of bias20 ,22 ,23) focused on the risk of herpes zoster infection. Comparing with csDMARDs, the aHRs varied between 1.0 and 1.7, most of them reflecting no statistically significant differences between the groups, especially when adjusted for drop-outs.20 ,22 ,24 Nevertheless, the data do not show the same direction and, therefore, a slightly increased risk cannot be ruled out. Compared with the general population, tuberculosis occurred more frequently in patients on TNFi, with a standardised incidence rate varying between 12 and 35.26 ,27
Regarding malignancies, a clear distinction needs to be made between studies having patients on csDMARDs and studies with the general population as the comparator. When comparing patients on TNFi with patients on csDMARDs, cancer does not occur more frequently (4 studies, 3 at low risk of bias,33 ,34 ,37 lymphoma does not occur more frequently (3 studies, 2 at low risk of bias34 ,39), NMSC does not occur more frequently (3 studies, 1 at low risk of bias42) and melanoma may occur more frequently (1 study at low risk of bias, aHR: 1.5 (95% CI 1.0 to 2.2)43). When comparing patients on TNFi with the general population, cancer does not occur more frequently (4 studies, 3 at low risk of bias33 ,34 ,37), lymphoma occurs more frequently (4 studies, 3 at low risk of bias34 ,39 ,40), with aHRs of 2.3–5.9, and NMSC may occur more frequently (1 study at low risk of bias, aHR: 1.7 (95% CI 1.4 to 2.0)42).
Five studies (3 at low risk of bias45 ,47 ,48) found no difference in mortality between patients on TNFi and patients on csDMARDs (see online supplementary tables S57–S62). For other safety outcomes, there is only data from few studies (see online supplementary tables S63–S98).
On average, patients on GCs had a higher risk of infections, which increased with higher dosage.12–18 It is unclear, however, if these increased infection rates should be attributed to GC use or to more severe RA, which in itself is associated with increased infection rate (confounding by indication).60 ,61 For low-dose GCs (in some studies defined as ≤5 mg and others as ≤7.5 mg), there was no increased risk, or the risk was only slightly increased (see online supplementary material, particularly tables S6, S13 and S80).
Totally, 6169 articles (after de-duplication) were screened and 12 were included (see flowchart in online supplementary figure S2). These included seven studies on infections (4 serious infections,12 ,14 ,16 ,18 2 skin infections21 ,24 and 1 tuberculosis62), 2 on cancer,43 ,63 1 on cardiovascular events,53 1 on neurological events51 and 1 on gastrointestinal perforation.56 Again, due to heterogeneity, a meta-analysis could not be performed. Details of the individual studies can be found in online supplementary tables S99–S138. For serious infections, from the 4 studies identified, 1 at low risk of bias, there were no difference in the risks across different csDMARDs.12 For other safety outcomes, there is only data from 1 to 2 studies. For tofacitinib, there is not yet data from observational studies.
With this SLR, we have investigated new safety signals of established sDMARDs and bDMARDS. We have confirmed the existing data that sDMARDs and bDMARDs are relatively safe for the treatment of RA. This does not mean that there are no risks at all: patients on TNFi, compared with patients on csDMARDs, have a higher risk of tuberculosis and a higher risk of serious infections, while a higher risk of herpes zoster cannot be excluded. In studies with low risk of bias, however, and when analyses were adjusted for drop-outs, results showed no risk for herpes zoster infection.20 ,22 ,24 Patients on TNFi do not have an increased risk of all types of malignancies, lymphoma or NMSC. The risk of melanoma may be increased in patients on TNFi, but these data come from one study only and need to be confirmed. Patients with a history of malignancy do not have an increased risk for another malignancy, but these data come from one study only, need to be confirmed, and should be considered with reserve.
In general, these conclusions, while stemming mainly from registries, confirm the findings from RCTs and from what was known from clinical practice.64 These conclusions are not ‘new’ and they were to some extent expected, but the source of the data is different, namely observational studies/registries, and furthermore, regular updates summarising the evidence are needed. In some instances, data from registries can give resolution in case of signals found in RCTs: a good example is the SLR on clinical trial data from Bongartz et al65 published in 2006 concluding that TNFi were associated with an increased risk of malignancies. Several studies from registries, with well-conducted analyses, have confirmed now that patients on TNFi do not have a higher risk of cancer.33–37 This reinforces the importance of registries to evaluate safety in comparison with RCTs. Among others, they have the advantage of including an unselected group of patients treated in routine care, who are followed-up for a long period of time.2 This makes the information on safety outcomes providing from registries of much more relevance for clinical practice. We have, therefore, opted for focusing this SLR on observational studies/registries.
The approach that was taken here assures that only the observational studies with highest methodological rigour are taken into consideration. Anecdotal information stemming from studies with less robust methodology may be excluded, thus explaining why this SLR only addresses general topics, such as the risk of malignancy or the risk of serious infection. Data on special situations, site-specific malignancies and infections is often lacking because of insufficient robust data.
Studies included in this SLR were very heterogeneous, precluding a meta-analysis. This substantial heterogeneity makes it difficult to draw solid conclusions when combining the evidence from all the studies. Furthermore, registries are different in their design, for instance, some are population-based and have access to other registries through linkage, while others are clinical-based and rely on data collection by the physician, which is more prone to missing events. It is therefore difficult to judge the completeness of the information of some of the registries, which may in the end lead to spurious results. Some of these problems could be overcome if studies reporting safety outcomes would follow the EULAR recommendations on analysing and reporting safety data of biologics registers.66 We have opted for a stringent methodology for this SLR and for only including studies that had a comparator group, so that the impact of the treatment on the occurrence of a safety event could be better interpreted. This is also what is recommended by the EULAR recommendations on analysing and reporting safety data of biologics registers. Because of this, studies from an important Japanese cohort were not included. These are notwithstanding impressive nation-wide postmarketing surveillance studies, covering thousands of patients, and also informative with respect to the safety profile of the different biologics.67–69
A note on non-TNFi biologicals, for which there is a lack of observational studies addressing safety aspects. Of the studies fulfilling the eligibility criteria of this SLR, only two included non-TNFi, but no separate results for non-TNFi are presented.55 ,56 Some initial registry data exist,70–74 and the safety profile from non-TNFi seems to be similar to that of TNFi. However, this is an extrapolation as these studies did not include a comparator group, and simply describe the occurrence of safety events in a group of patients under a given non-TNFi, reason for which they were not included in this SLR. Similarly, data from observational studies on the newer TNFi, golimumab and certolizumab pegol, or biosimilars are not yet available. Registries have shown to be a powerful tool to give us relevant information especially in what concerns safety of drugs. Nevertheless, several crucial questions remain largely unanswered, such as: what is the true risk of a bDMARD in patients with a previous malignancy? Is it really safe to start rituximab in patients with history of a previous malignancy? Is it safer to start a biologic with a different mechanism of action in case of a malignancy occurring during or after a bDMARD? And so on. These important questions should be elaborated in future studies.
No new signals were found for csDMARDs, but, despite a few exceptions, no high-quality studies were captured either. Actually, the methodology that has been implemented for the biologic registries has marked a new era in outcomes research in rheumatology. It would be desirable if these efforts were used not only to study these new-generation drugs, but also to get more insight into drugs that are widely used, such as csDMARDs and GCs. Rheumatologists are convinced of the efficacy of GCs, give them a prominent place in the treatment algorithms of RA, but still fear their safety profile. In part, this is due to the lack of robust safety data on GC, and this may be addressed in registries. With respect to tofacitinib, information on safety can so far only be obtained from RCTs, which is limited. Some concerns on the safety profile of this drug have been raised, namely the risk for herpes zoster infection.75 Accumulating experience with tofacitinib will, in the near future, also offer the possibility of having real-life data from patients on this drug, from which we will learn more about its safety profile, among others, about the real importance of infection by herpes zoster in patients treated with tofacitinib.
The authors thank Louise Falzon, librarian at the Center for Behavioral Cardiovascular Health, Columbia University Medical Center, New York and former Trial Search Coordinator from the Cochrane Musculoskeletal Group, for assisting with the design of the search strategy.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Handling editor Francis Berenbaum
Contributors All authors contributed and finally approved the current manuscript.
Competing interests Conflicts of interest are summarised below as remuneration for consultation and/or speaking engagements (R) and/or research funding (F). SR—R: Fundação para a Ciência e Tecnologia. CG-V—R: Abbvie, BMS, MSD, Pfizer, Roche-Chugai, UCB; F: Expanscience, Nordic Pharma, Pfizer. JLN—R: UCB. JSS—R: Abbott/Abbvie, Amgen, Astra-Zeneka, BMS, Celgene, Glaxo, Infinity, Janssen, Lilly, Medimmune, Menarini, MSD, Novo-Nordisk, Pfizer, Roche, Samsung, Sandoz, Sanofi-Aventis, UCB, Vertex; F: Abbott, BMS, MSD, Pfizer, Roche, UCB. MB—R: Abbott, Bristol Myers-Squibb, Chugai, Pfizer, Roche; F: Pfizer. LG—R: Abbott, BMS, Chugai, Pfizer, Roche, UCB. DvdH—R: Abbott, Amgen, AstraZeneca, BMS, Centocor, Daiichi, Eli-Lilly, GSK, Janssen Biologics, Merck, Novartis, Novo-Nordisk, Otsuka, Pfizer, Roche, Sanofi-Aventis, UCB; Director of Imaging Rheumatology bv; F: UCB, Pfizer. KW—R: Pfizer, Genentech, UCB, Abbott. RL—R: Abbott/AbbVie, Ablynx, Amgen, Astra-Zeneca, Bristol Myers Squibb, Centocor, Glaxo-Smith-Kline, Novartis, Merck, Pfizer, Roche, Schering-Plough, UCB, Wyeth; F: Abbott, Amgen, Centocor, Novartis, Pfizer, Roche, Schering-Plough, UCB, Wyeth.
Provenance and peer review Not commissioned; externally peer reviewed.