Objectives To investigate whether tumour necrosis factor alpha inhibitors (TNFis) are associated with an increased risk of neuroinflammatory diseases among patients with arthritic diseases.
Methods Cohorts of patients with rheumatoid arthritis (RA, n=25 796), psoriatic arthritis (PsA, n=8586) and ankylosing spondylitis (AS, n=9527) who initiated a TNFi treatment year 2000–2017 were identified from nationwide clinical rheumatology registers in Sweden and Denmark. Information on demyelinating disease and inflammatory neuropathy diagnoses was retrieved from prospective linkage to National Patients Register. A Cox proportional hazard model was used to estimate HRs and 95% CI comparing TNFi exposed and non-exposed, by disease and country.
Results Among 111 455 patients with RA, we identified 270 (Sweden) and 51 (Denmark) events (all types of neuroinflammatory diseases combined), corresponding to crude incidence rates (per 1000 person-years) of 0.37 (Sweden) and 0.39 (Denmark) in TNFi-treated patients vs 0.39 (Sweden) and 0.28 (Denmark) in unexposed patients, and an age-sex-calendar-period-adjusted HR (95% CI) of 0.97 (0.72 to 1.33) (Sweden) and 1.45 (0.74 to 2.81) (Denmark) in TNFi exposed compared with non-exposed patients. For a total of 64 065 AS/PsA patients, the corresponding numbers were: 196 and 32 events, crude incidence rates of 0.59 and 0.87 in TNFi-treated patients vs 0.40 and 0.19 in unexposed patients, and HRs of 1.50 (1.07 to 2.11) and 3.41 (1.30 to 8.96), for Sweden and Denmark, respectively. For multiple sclerosis, the patterns of HRs were similar.
Conclusions Use of TNFi in AS/PsA, but not in RA, was associated with increased risk of incident neuroinflammatory disease, though the absolute risk was below one in 1000 patients/year.
- rheumatoid arthritis
- psoriatic arthritis
- ankylosing spondylitis
- outcomes research
Statistics from Altmetric.com
What is already known about this subject?
Numerous case reports suggest that multiple sclerosis-related disorders may be adverse events following treatment with tumour necrosis factor alpha inhibitor (TNFi) for arthritic diseases.
What does this study add?
This study included 175 520 arthritic patients of whom 43 909 were treated with TNFi in routine care in Denmark and Sweden and followed for up to 17 years for neuroinflammatory events in National Patient Registers.
An increased but rare risk of neuroinflammatory events following TNFi treatment among patients with psoriatic arthritis and ankylosing spondylitis compared with non-treated patients was observed in both Sweden and Denmark.
No consistent and significant risk of neuroinflammatory events following TNFi treatment among patients with rheumatoid arthritis was found.
How might this impact on clinical practice or future developments?
The risk profile following treatment with TNFis may be different for inflammatory arthritides, which has impact on decision making in clinical practice.
Tumour necrosis factor alpha inhibitors (TNFis) have successfully established themselves as mainstay therapies for rheumatoid arthritis (RA), psoriatic arthritis (PsA) and ankylosing spondylitis (AS). Several case reports and small case series have indicated that treatment with TNFi treatment may be linked with neuroinflammatory disorders, for example, multiple sclerosis (MS), inflammatory neuropathies and optic neuritis.1–11
Indeed, available evidence suggests an important role of TNF in the pathogenesis of MS. High levels of TNF have been found in cerebral white matter lesions suggestive of MS and cerebrospinal fluid of patients with MS12 13; and a genetic variant in the gene encoding TNF-receptor 1 confers susceptibility to MS.14 Patients with MS treated with TNFi in randomised clinical trials have shown unfavourable results or have even been halted due to dose-dependent disease exacerbation.15 16 The background risk of developing a neuroinflammatory event of these types may, however, depend on the type of arthritides. For example (and in the absence of TNFi treatment), PsA patients have been reported to be at increased risk for MS,17 while an inverse association has been suggested for RA and MS.18
To examine whether such neuroinflammatory events are merely coincidental or causally linked to the use of TNFi, and how any association may vary by treatment indication, large observational studies with sufficiently long follow-up are warranted. To date, only a few observational studies exist, mostly pointing towards increased risk of neuroinflammatory adverse events following TNFi treatment.19–21 We previously reported an association between TNFi treatment and MS among men (but not women) with RA and AS compared with the general population.20 However, that study had limited power, nor did it capture other MS-related conditions, such as other types of demyelination or optic neuritis.
The purpose of the current study was, therefore, to investigate, in a much larger cohort, the occurrence of MS and other types of neuroinflammatory events, and to assess risks specifically by TNFi treatment and by rheumatological condition. To do this, we combined nationwide clinical and health registers from Sweden and Denmark.
Materials and methods
We performed a prospective cohort study in Denmark and Sweden in parallel (ie, all analyses were run separately in each country), with inflammatory arthritis patients identified within national registers. DANBIO is a Danish nationwide quality and research register that covers 90%–98% of adults with rheumatic diseases treated with TNFi and other biologic disease-modifying anti-rheumatic drugs (bDMARDs) in routine care.22 23 The Swedish Anti-Rheumatic Treatment in Sweden Register (ARTIS) is a subset of the nationwide Swedish Rheumatology Quality Register that covers 87%–95% of bDMARD-treated patients with rheumatic diseases.24 In both registers, clinical data are collected prospectively. Diagnoses are validated by yearly medical record audits (DANBIO) with a positive predictive value of 96% for RA diagnoses in 201525 and the quality and coverage of the ARTIS register is regularly checked through reports and scientific publications.26 The nationwide Danish and Swedish National Patients Registers (NPR) were used to identify outcomes (selected neuroinflammatory diagnoses), and for identifying Swedish arthritis patients with RA, AS or PsA who were unexposed to TNFi. These registers are administrative and funded through public taxation and contain nearly complete information about inpatient and outpatient care.27 28 Finally, the Civil Registration System29 and the Swedish Population Register28 were used for retrieving information on dates of birth, death or emigration, respectively. The personal identification number assigned to every Danish and Swedish citizen permitted register linkage.
Cohort identification and follow-up
The Danish data included patients from DANBIO with a diagnosis of RA, PsA or AS between 1 January 2000 and 20 January 2017. The Swedish data included patients with a diagnosis of RA, PsA or AS between 1 January 2000 and 29 December, 2017 identified from the NPR and/or from ARTIS (for all practical purposes, this setting could be seen as an identification of all TNFi exposed patients from the national underlying source population of all RA, PsA or AS). Patients were at least 18 years old at start of follow-up in both countries.
Exposure assessment and start of follow-up
In both countries, follow-up for patients exposed to TNFi (originator or biosimilar) started at the date of first TNFi treatment. For patients who were (yet) naïve to TNFi start of follow-up was defined as the first registered visit in DANBIO (Denmark), and for patients in Sweden, start of follow-up was at the second visit listing (RA or PsA/AS) in the NPR, or 1 January 2000, whichever came last, and with the additional condition that at least one of the two required arthritis diagnoses was established in a hospital Department of Rheumatology or Internal Medicine, a validated method to capture rheumatological diagnoses in NPRs.25 30 31 We did not consider use of non-TNFi biologicals which was used in less than 10% of patients in both cohorts.
Outcome information and end of follow-up
In addition to MS, we searched for a broader array of diagnostic codes that may correspond to neuroimmune conditions occurring as a result of TNFi treatment1–11 using the WHO International Classification of Diseases 10th revision, which are coded by the discharging physician when a patient is discharged from hospital (table 1). All registered neuroinflammatory events were combined and categorised in three subgroups: (1) demyelinating diseases of the central nervous system including optic neuritis, subsequently termed ‘demyelinating diseases’; (2) MS and (3) inflammatory polyneuropathies, subsequently termed ‘polyneuropathies’. Patients with one of the above-mentioned neuroinflammatory diagnoses prior to start of follow-up were excluded from the cohorts. The validity of MS in the Danish NPR has been found to be 96.3%.32
The cohorts were followed up from entry date until the date of the first registered neuroinflammatory diagnosis, death, emigration or the end of study (DANBIO: 10 May 2017 and Sweden: 31 December 2017), whichever came first. TNF-naïve individuals who, during follow-up, started a TNFi treatment were censored from the TNFi-naïve group and contributed person-time and any events to the TNFi exposed group thereafter.
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
Baseline demographic and clinical characteristics of the cohort at entry are presented as numbers with their percentages and medians with IQR. Due to the prospective design, patients could contribute to both the exposed and the unexposed groups. In that case, baseline demographic and clinical characteristics were retrieved at the two time points corresponding to the start of follow-up of the episode in question.
A Cox proportional hazards model, with attained age as time scale, was used to estimate HRs and 95% CIs for the comparison of TNFi treated to non-treated patients with delayed entry. Adjusted models included gender and year of inclusion in the cohorts (categorised as: 2000–2005, 2006–2011 and 2012–2017). Exposure to TNFi was included in the model as a time-dependent variable. The proportionality assumption of the Cox proportional hazard model was checked for all variables using the method by Lin et al.33
The following sensitivity or subanalyses were performed: (1) An ‘on drug’ analysis, in which follow-up was censored at the date of discontinuation of the TNFi treatment (plus a 3-month interval), (2) Stratification by gender, (3) Stratification by follow-up time (<1 year, 1–4 years and ≥5 years) and (4) Stratification by age at treatment start or cohort entry (age <50, 50–64 and ≥65 years). In addition, for the DANBIO cohort only, a subanalysis was made with adjustment for smoking status (entered categorically as ‘previous’, ‘current’ or ‘never’ smokers at time of the first visit in DANBIO), and, among RA patients only, Disease Activity Score using 28 joint count and C reactive protein (DAS28-CRP), when this information was recorded (complete-case analysis). Finally, for the Swedish cohort only, we performed a sensitivity analysis that employed alternative definitions of the unexposed comparator: (1) the unexposed patients identified from ARTIS (ie, using the TNFi naïve patients from the Rheumatology Register instead of the NPR); (2) the unexposed patients from NPR restricted to patients treated with a conventional synthetic (cs) DMARD who either switched or escalated to another csDMARD, the date of switching taken as the follow-up start, that is, a new-user active comparator approach. The study period for the latter analysis was restricted to 2006–2017 due to the availability of the Swedish Prescribed Drug Register from which csDMARD information was retrieved34; (3) the unexposed patients from NPR with a start of follow-up at the second visit listing (RA or PsA/AS) in the NPR, or 1 January 2006, whichever came last (study period 2006–2017) and (4) same as the latter one with 2011 instead of 2006 (study period 2011–2017).
The number needed to harm (NNH) was calculated as the inverse of the crude excess incidence (ie, crude incidence rateexposed – crude incidence ratenon-exposed).
A p<0.05 was considered significant for all tests. Statistical analyses were conducted using SAS V.9.4 (SAS).
We included a total of 175 520 patients with RA, PsA or AS in the two cohorts of whom 43 909 (25%) patients were exposed to TNFi any time during follow-up (table 2).
The overall proportion of TNFi exposed patients varied from 23% among patients with RA in the Swedish cohort to 38% among patients with PsA and AS in DANBIO. Median follow-up time varied between 2.7 years for non-exposed PsA and AS patients in DANBIO to 7.4 years among RA patients exposed to TNFi in both cohorts. Generally, TNFi exposed patients were younger at entry and at time of diagnosis than non-treated patients in both countries, they had a longer disease duration at entry, and they had higher disease activity based on clinical and subjective measures.
For Sweden and Denmark, respectively, we identified the following number of events per person-years: 160 events/706 386 person-years and 33 events/162 003 person-years for ‘demyelinating diseases’, 49 events/706 830 person-years and 11 events/162 118 person-years for MS; and 110 events/706 650 person-years and 18 events/162 088 person-years for ‘polyneuropathies’. Taking all neuroinflammatory events combined in each country, the crude incident rates (per 1000 person-years) across all patients exposed to TNFi were 0.37 (Sweden) and 0.39 (Denmark) vs 0.39 (Sweden) and 0.28 (Denmark) in unexposed patients (table 3).
We did not find any association between treatment with TNFi and risk of a neuroinflammatory event, (HR 0.97, (95% CI 0.72 to 1.33)) for the Swedish cohort and an elevated but not significantly increased (HR 1.45, (95% CI 0.74 to 2.81)) in the Danish cohort (table 3). For MS, a borderline statistically significantly reduced risk after exposure to TNFi was found in Sweden (HR 0.44 (95% CI 0.18 to 1.05)). In Denmark, the number of MS events was very low (four events among the exposed vs seven events among the unexposed), hence, it was not possible to confirm or refute the Swedish results.
Among TNFi exposed patients experiencing an event (of any type), the median time to event counting from start of TNFi treatment was 3.6 years (IQR=1.1–6.8) and 3.2 years (IQR=1.4–5.1) in the Swedish and Danish cohorts, respectively (online supplementary table 1).
Sensitivity and subanalyses
Our sensitivity analyses did not demonstrate any signal of TNFi treatment being associated with risk of a neuroinflammatory event (all types combined) (online supplementary table 2). In both cohorts, though, a tendency towards increased risk of a neuroinflammatory events was seen with increasing age.
AS and PsA
In Sweden and Denmark, we identified the following number of events per person-years: 141/447 906 person-years and 24/60 835 person-years events of ‘demyelinating diseases’, 48/448 324 person-years and 7/60 894 person-years MS events, 55/448 497 person-years and 8/60 891 person-years polyneuropathy events, respectively. Taking all neuroinflammatory events combined in each country, the crude incidence rates (per 1000 person-years) in patients exposed to TNFi were 0.59 (Sweden) and 0.87 (Denmark) vs 0.40 (Sweden) and 0.19 (Denmark) in unexposed patients. These rates translated into a 50% increased risk of a neuroinflammatory event (all types combined) among patients ever exposed to TNFi in the Swedish cohort (HR 1.50 (95% CI 1.07 to 2.11)) and a 3.4-fold increased risk in the Danish cohort (HR 3.41 (95% CI 1.30 to 8.96)) compared with patients not exposed to TNFi (table 3). This corresponds to an NNH of 5300 patients treated with TNFi/year in Sweden and 1500 patients in Denmark.
Among TNFi exposed patients experiencing an event, the time to (any type of) neuroinflammatory event was 3.8 years (IQR=1.7–5.9) and 3.1 years (IQR=1.4–8.6), and age at neuroinflammatory event among TNFi exposed patients was 43 years (IQR=34–52) and 46 years (IQR=38–55) in the Swedish and Danish cohort, respectively (online supplementary table 1).
Sensitivity and subanalyses
The ‘on-drug’ analyses’ estimates were in agreement with the results of the main analysis although the risk estimates were lower and less precise in both countries (online supplementary table 2). The HR for male patients was twice the HR for female patients in the Danish cohort (5.81 vs 2.50), while in the Swedish cohort, the HRs were of the same order of magnitude in both sexes (1.37 for males vs 1.67 for females). However, despite the difference in HRs between males and females in the Danish cohort, gender did not appreciably modify the HR for the association between TNFi and the outcome(s) under study (ie, no interaction was found). The risk for a neuroinflammatory event was increased with increasing follow-up time in the Swedish cohort, whereas this was less evident in the Danish cohort due to low number of events in each stratum. No apparent association with age at cohort entry or treatments start was found. When restricting the unexposed patients in Sweden to csDMARD switchers or when the unexposed patients were retrieved from the Swedish Rheumatology Register instead of NPR, a doubled risk of neuroinflammatory event was seen for TNFi exposed patients compared with non-exposed patients (online supplementary table 2). Crude incidence rates were similar in AS and PsA patients in the Danish cohort. In the Swedish cohort, the incidence rates were higher in patients with AS than in PsA, but the ratios between exposed and unexposed were of the same magnitude in both diseases (online supplementary table 3).
In this large population-based, prospective cohort study, we observed an increased risk of neuroinflammatory events suggestive of demyelination following treatment with TNFi among patients with PsA or AS compared with non-exposed PsA or AS patients although the risk was very low. We found no significantly increased risk of neuroinflammatory events among RA patients treated with TNFi compared with non-treated RA patients, suggesting that the observed safety profile of TNFi with respect to MS and similar conditions varies with the treated diagnosis.
Two other studies have assessed the risk of different neuroinflammatory events following TNFi treatment in arthritis patients. Both studies used US health claim databases and a nested case–control design. The study by Bernatsky et al found a non-significant 31% increased risk of demyelinating events following TNFi exposure among RA patients compared with non-exposed RA patients,19 whereas Etminan et al found a more than twofold increased risk of peripheral neuropathy among arthritis patients (the study combined RA, PsA and AS into one group) with a past use of TNFi compared with arthritis patients not exposed to TNFi.21 Both studies had relatively short mean follow-up times (1.9 and 2.3 person-years, respectively) and the claims-based database used where unrepresentative, based of unemployed and older individuals. However, this might suggest an increased risk in the elderly patients as corresponds with our findings of a trend of increased risk with age among RA patients, although not significant.
Our large population-based study had a longer follow-up time than these studies and demonstrate that the pattern of risks for developing a neuroinflammatory event with TNFi treatment differs between RA and PsA/AS patients. Since chronic inflammatory diseases have distinct genetic profiles35 36 and different distribution of age and sex, these diseases should be separated in the risk analyses.
The underlying mechanisms of demyelinating disorders and inflammatory polyneuropathies following TNFi treatment that have been reported in the literature (mostly in case reports and small case series), are not completely understood, but several mechanisms have been proposed including the lack of entry through the blood–brain barrier theory, aggravation of CNS demyelination by decreasing TNF receptors and unfavourable altering of cytokine responses.37
The risk estimates obtained from the ‘on-drug’ analysis among PsA and AS patients were lower than in the main analysis which may suggest that demyelinating events develop after longer exposure time, and even after the treatment has been discontinued. However, in our study, and due to data availability, the time to event varied across cohorts, thus making it difficult to confirm the nature of any temporal association between exposure and outcome. However, our results are in keeping with a previous study from Denmark that observed six patients with RA, PsA or AS with a demyelinating neuroinflammatory disorder following TNFi treatment, observed from 5 months to 4 years after treatment start.5
There are limitations to the study. First, confounding by indication, or channelling, may explain some of our associations. This risk has, however, been mitigated by excluding patients who registered with any neuroinflammatory outcome prior to cohort entry. TNFi exposed RA patients had a more active disease based on clinical and objective disease activity measures. However, we do not know whether disease activity is associated with the risk of having a demyelinating disease or inflammatory neuropathy—and the risk estimates did not change considerably when adjusting for DAS28-CRP in subanalysis. Although validation studies have found high positive predictive values for RA, AS and PsA (79%–96%), there may be differences in the validity of treated vs untreated disease, though we do not believe these differences are large enough to explain our findings. Further, we cannot exclude residual or unknown confounding.
The strength of the study is the use of high-quality population-based rheumatic disease registers with long follow-up times (up to 17 years) supplemented with near-complete nationwide patient registers. The use of data from two countries not only increased statistical power, but also provided a built-in means to compare patterns in the two countries, and thus the generalisability of the results. We included patients with the same underlying disease as comparison group which reduced the risk of surveillance bias.38 Another feature of the current study is the agreement between the results from the two countries: despite slightly different definitions of the TNFi-naïve comparator cohorts in the two countries, the results point to joined conclusions. Since defining an unexposed group that is fully comparable to the TNFi exposed group is a real challenge in arthritides patients, a series of alternative comparator groups were defined for sensitivity analyses. All results pointed to similar conclusions, which strengthens the findings.
In conclusion, this large prospective cohort study including 175 520 patients followed for 1 376 149 person-years showed that the use of TNFi for the treatment of PsA or AS, but not for RA, may be associated with increased risk of having a demyelinating disease or inflammatory neuropathy. This information will be important for risk communication and evaluation in clinical practice, even though the absolute risk is low. The underlying biological mechanism needs to be further explored to characterise the mechanism of action and to enable identification of susceptible patients.
We thank the DANBIO and ARTIS registers for allowing us the use of the clinical data from their patients. Hospital Departments of Rheumatology and the private rheumatologic clinics in Denmark and Sweden are acknowledged for their contribution to DANBIO and ARTIS.
Handling editor Josef S Smolen
TIK and BD contributed equally.
Contributors LD and JA conceived the idea for the study and all authors planned the study collaboratively. TIK, BD, EVA, RKJ, JA and LD developed the analysis plan. TIK, BD, EVA and RKJ performed management of data and statistical analyses. MM and FS provided specific input on neuroinflammatory diseases. HL, JA and LD provided specific input on rheumatic diseases. RLC and DVJ provided input on DANBIO data. JA and LD provided funding. TIK and BD wrote the first draft of the manuscript. All authors reviewed the results and critically reviewed the manuscript for intellectual content. All authors approved the final version of the manuscript.
Funding This study was funded by Program for Clinical Research Infrastructure (PROCRIN) established by the Novo Nordisk and Lundbeck Foundations, and received funding from the Swedish Research Council, The independent Research Fund Denmark, and from the Karolinska Institutet Region Stockholm funds (ALF).
Competing interests MM has received grants from Novartis and Biogen, and has received personal fees from Novartis, Biogen, Merck, Sanofi Genzyme and Teva. FS has received grants and personal fees from Biogen, Merck, Novartis, Sanofi Genzyme and Roche. JA has received grants from Abbvie, BMS, Eli Lilly, MSD, Pfizer, Roche, Samsung Bioepis, Sanofi and UCB. LD has received grants from BMS and has received personal fees from Eli Lilly and Galderma. TIK, BD, EVA, RKJ, EHI, RLC, HL and DVJ have nothing to disclose.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Patient consent for publication Not required.
Ethics approval According to Danish and Swedish legislation, the registration and publication of data from national registers do not require patient consent or approval by Ethics Committees. Approval was given by the Danish Data Protection Agency (j.no: 04422, with I-Suite no: CSU-FCFS-2016-001).
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement No data are available.
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.