Malignant Neoplasms in Patients With Rheumatoid Arthritis Treated With Tumor Necrosis Factor Inhibitors, Tocilizumab, Abatacept, or Rituximab in Clinical Practice: A Nationwide Cohort Study From Sweden

Key Points

Question  Does treatment with tocilizumab, abatacept, rituximab, or tumor necrosis factor (TNF) inhibitors affect the risk of malignant neoplasms among patients with rheumatoid arthritis?

Findings  In this register-based cohort study, the risk of malignant neoplasms did not differ between patients treated with a first anti-TNF drug, a second anti-TNF drug, tocilizumab, abatacept, rituximab, or conventional synthetic disease-modifying antirheumatic drugs, with the possible exception of an increased risk of squamous cell skin cancer risk in patients treated with abatacept.

Meaning  Short- to medium-term use of tocilizumab, abatacept, rituximab, or anti-TNF drugs seems to be safe with regard to risks of incident cancer.

Importance  Considering the widespread and increasing use of biological immunomodulators (biological disease-modifying antirheumatic drugs [bDMARDs]) to treat chronic inflammatory conditions, and the concern that immunomodulation may alter cancer risk and progression, the limited available data on use of these therapies as used in clinical practice and cancer risks are a concern.

Objective  To assess the risk of incident malignant neoplasms in patients with rheumatoid arthritis (RA) treated with bDMARDs.

Design, Setting, and Participants  This was a national register–based prospective cohort study of the public health care system in Sweden from 2006 to 2015. Cohorts of patients with RA initiating treatment with tocilizumab (n = 1798), abatacept (n = 2021), and rituximab (n = 3586), a tumor necrosis factor inhibitor (TNFi) as first-ever (n = 10 782) or second-ever (n = 4347) bDMARD, a biologics-naive cohort treated with conventional synthetic disease-modifying antirheumatic drugs (csDMARDs) (n = 46 610), and a general population comparator cohort (n = 107 491).

Exposures  Treatment with tocilizumab, abatacept, rituximab, or TNFi.

Main Outcomes and Measures  Outcomes included a first invasive solid or hematologic malignant neoplasm, or skin cancer. Hazard ratios were calculated using Cox-regression, adjusted for age, sex, disease and treatment characteristics, and educational level.

Results  We identified a total of 15 129 initiations of TNFi as the first or second bDMARD, 7405 initiations of other bDMARDs, and 46 610 csDMARD users. The mean age varied from 58 to 64 years, and the proportion of female patients varied from 71% to 80%, across the 7 cohorts under study. The observed numbers of events (crude incidence per 100 000 person-years) for a first invasive solid or hematologic malignant neoplasm were 50 (959) for tocilizumab, 61 (1026) for abatacept, 141 (1074) for rituximab, 478 (978) for initiators of TNFi as first bDMARD, and 169 (917) for TNFi as second bDMARD. There were no statistically significant differences between initiators of a first or second TNFi, or other bDMARDs, and bDMARD-naive RA for any of a total of 25 drug- and outcome-specific comparisons, with 1 exception (abatacept and increased risk of squamous cell skin cancer).

Conclusions and Relevance  The overall risk of cancer among patients with RA initiating TNFi as first or second bDMARD, tocilizumab, abatacept, or rituximab does not differ substantially from that of biologic drug–naive, csDMARD-treated patients with RA, although altered risks for specific cancer types, or those with longer latency, cannot be excluded.

Currently available biologic disease-modifying antirheumatic drugs (bDMARDs) for rheumatoid arthritis (RA) include agents targeting tumor necrosis factor (TNF) (adalimumab, etanercept, certolizumab, golimumab, infliximab), CD20 (rituximab), CD80/CD86 via CTLA-4 (abatacept), and interleukin 6 (tocilizumab). Immune incompetence may lower host surveillance against incipient tumors and accelerate cancer progression.^1,2 Conversely, for several malignant neoplasms, immune therapies have emerged as a promising therapeutic approach, in part targeting the same pathways. For instance, abatacept is a CTLA-4 fusion protein used in the treatment of RA. Ipilimumab, an antibody that blocks CTLA-4, is approved for the treatment of malignant melanoma. Thus, the occurrence of malignant neoplasms in patients with chronic inflammatory diseases treated with immunomodulatory drugs needs to be carefully assessed.

Reports on TNF inhibitors (TNFi) as used in RA and risk of malignant neoplasms have mostly been reassuring.^3-9 The patient populations in these studies have, however, been reflective of those patients who were among the first to receive TNFi treatment more than a decade ago,^8-12 and whose disease characteristics and other treatment exposures differ from contemporary patients, resulting in a need for updated assessments.

The risk of malignant neoplasms in RA treated with the non-TNFi bDMARDs (non-TNFi), rituximab, abatacept, and tocilizumab, is considerably less investigated. Pooled data from randomized clinical trials and open-label extension studies have not provided any signal of increased risk for cancer overall.^13-19 However, trial exclusion criteria and pretrial screening procedures reduce the generalizability of trial data because patients at high risk of developing cancer may not be represented. The typical use of cancer data from the general population, or external RA cohorts, as benchmark^14,20,21 in open-label extensions is problematic because the observed cancer incidence may be modified by effects related to the treated disease itself, starting a new therapy, and participation in a trial (extension).^22,23

We therefore aimed to assess cancer risks in patients with RA treated with TNFi or other bDMARDs in contemporary clinical practice, compared with bDMARD-naive patients with RA, and to contextualize these risks using the general population.

We performed a nationwide cohort study of patients with RA starting TNFi treatment as their first and second bDMARD; patients with RA starting treatment with non-TNFi; bDMARD-naive patients with RA initiating, or stable while taking, respectively, conventional synthetic disease-modifying antirheumatic drugs (csDMARDs); and matched general population comparator subjects. All bDMARD cohorts were followed from treatment start.²³ The study was approved by the ethics committee at the Karolinska Institutet. Patients did not provide written informed consent because this was a register-based study that, according to Swedish law, does not require individual written informed consent.

Swedish health care is public and tax funded. Prescription drugs are provided free of charge above a threshold of $250 USD annually. Most patients with RA are treated by, and most outpatient rheumatology care is provided by, hospital-based rheumatologists. Most patients who start treatment with non-TNFi have previously been treated with TNFi. Sweden has national and virtually complete registers on demographics and health care data that can be linked together by the unique personal identity number issued to all Swedish residents.

The Swedish Rheumatology Quality of Care Register (SRQ) is a profession-based register that covers 90% to 95% of all patients with RA treated with bDMARDs in Sweden.^24,25 The nationwide Swedish Patient Register lists diagnoses, dates, and hospital and department. Diagnoses are listed as assigned by the discharging physician; the inpatient subset has had near-complete coverage since 1987,²⁶ the outpatient subset since 2001. Validations against medical files suggest a positive predictive value for the RA diagnosis of around 90%.²⁷ Reporting of incident cancers to the Swedish Cancer Register is mandatory, the coverage is estimated to be more than 95%.²⁸ The register contains data on date of diagnosis, type of cancer according to the International Classification of Diseases, Revisions 7 to 10, codes, and morphologic and histologic findings. The Prescribed Drug Register is a nationwide register with complete coverage that has gathered data on dispensations of prescribed drugs in Sweden since 2005. The Total Population Register lists data on residency, including dates of emigration and immigration. The Causes of Death Register lists date of deaths.

In the Patient Register, we identified all individuals with 2 or more visits in nonprimary outpatient care listing a main or contributory diagnosis of RA from 2006 to 2015. At least 1 of these visits had to be at a rheumatology or internal medicine department.²⁷ The second of the 2 visits served as date of inclusion. Patients who also received a diagnosis of juvenile idiopathic arthritis, systemic lupus erythematosus, psoriatic arthritis, or a spondyloarthropathy at a rheumatology or internal medicine department prior to or during follow-up were excluded or censored, respectively. Through linkage of this RA cohort to SRQ/Anti-Rheumatic Therapy in Sweden (ARTIS) and the Prescribed Drug Register, all bDMARD-initiations from 1999 to 2015 in patients with RA were identified. These linkages identified patients, who during our study period (2006-2015), initiated treatment with a TNFi as their first (n = 10 782) or second (n = 4347) bDMARD, patients who started treatment with non-TNFi (tocilizumab, n = 1798; abatacept, n = 2021; rituximab, n = 3586) during our study period regardless of previous bDMARD exposure, and patients who were bDMARD naive but initiated treatment, or received ongoing treatment at baseline, with at least 1 csDMARD (leflunomide, azathioprine, methotrexate, cyclosporine, antimalarial drugs, sulfasalazine, or gold) according to the Prescribed Drug Register (n = 46 610). Through linkage of the bDMARD-treated cohorts to the Population Register, we assembled a general population comparator cohort (individually matched 1:5 to the bDMARD cohorts by year of birth, sex, and area of residence [n = 107 491]).

Start of follow-up for the bDMARD cohorts was the date of the cohort-defining treatment start during 2006 to 2015. Start of follow-up for the bDMARD-naive cohort was defined as the first date when all the diagnosis- and DMARD-related criteria were fulfilled from 2006 to 2015. The inclusion date for the general population cohort was set to that of their corresponding bDMARD-treated patients. In the main analysis, a “once exposed, always exposed” approach was used. An “on-drug” approach was also performed; the bDMARD-naive patients were censored at start of the first bDMARD, and the bDMARD cohorts were censored 90 days (180 days for patients treated with rituximab) after therapy discontinuation. End of follow-up was defined as December 31, 2015; death; last emigration from Sweden; date of any solid organ transplantation; or occurrence of the outcome under study, whichever came first. For each outcome and cohort, patients who had previously developed the outcome under study were excluded.

The following 5 outcomes were assessed separately: (1) first invasive solid or hematologic malignant neoplasm, excluding nonmelanoma skin cancer (NMSC); (2) first invasive solid malignant neoplasm, excluding NMSC; (3) first invasive hematologic malignant neoplasm; (4) first invasive squamous cell skin cancer; and (5) first invasive melanoma.

Two alternative definitions of the bDMARD-naive cohort were assessed, one that included all bDMARD-naive patients, with no csDMARD requirement (“bionaive RA”), and another in which patients with RA receiving any ongoing csDMARD treatment were followed-up from the first date of switching to or adding a new csDMARD (“csDMARD switchers”). Also, to reduce any detection bias in conjunction with start of bDMARD therapy, we added a 3-month lag period between inclusion and start of follow-up.

Data on educational level were retrieved from Statistics Sweden and defined as less than 9, 9 to 12, and more than 12 years of education. Through linkage of all individuals with RA to the Swedish Patient Register, all outpatient records and hospitalizations were identified. Ever having been diagnosed as having ischemic heart disease; chronic obstructive pulmonary disease; diabetes mellitus); and knee, ankle, hip, or shoulder joint surgery was included as a proxy for general health and RA disease severity, both of which may be linked to cancer incidence and the likelihood of having an incipient cancer detected. In addition, overall health care utilization was measured as number of hospitalizations and total days spent in hospital since 1987. Data on sick leave and disability pension for the full calendar year before start of follow-up were collected. Use of oral prednisolone, csDMARDs, and nonsteroidal anti-inflammatory drugs at start of follow-up was determined (yes/no) by recorded use within 3 months in the Prescribed Drug Register. In addition, for the bDMARD cohorts information on rheumatoid factor,²⁹ Health Assessment Questionnaire (HAQ), Disease Activity Score 28 C-reactive protein (DAS28-CRP), CRP level, erythrocyte sedimentation rate (ESR) (both categorized according to quartiles with missing values included as a fifth category), and RA disease duration (<5, 5-10, and >10 years) was collected from the SRQ. All covariates reflected status at start of follow-up, in each cohort and for each model.

Number of events, person-years at risk, and crude- and age-standardized incidences were calculated for each outcome. Hazard ratios (HRs) were calculated using Cox proportional hazards model with time since treatment start as timescale. All HRs were adjusted for age, sex, and start year. Hazard ratios comparing bDMARD and csDMARD cohorts were further adjusted for educational level, comorbidities, number of hospitalizations and days spent in hospital care (1987-present), use of prednisolone at baseline, use of NSAID at baseline, number of prescription drugs at baseline, and sick leave and disability pension (none vs any) the year before cohort entry. Hazard ratios comparing different bDMARD cohorts were additionally adjusted for prior bDMARD therapy, rheumatoid factor, DAS28-CRP, HAQ, ESR, CRP level, and RA disease duration. Hazard ratios comparing csDMARD-treated RA and the general population were, in addition to age, sex, and start year, adjusted for area of residence. Hazard ratios and standardized incidences were calculated only if the smallest cell contained 5 or more events. We used SAS statistical software (version 9.4; SAS Institute Inc).

We identified a total of 15 129 initiations of TNFi as the first or second bDMARD, 6358 initiations of non-TNFi, and 46 610 csDMARD users. Baseline characteristics differed across the study cohorts (Table 1).

For initiators of a first-ever TNFi, the rate of an invasive solid or hematologic malignant neoplasm was 978 per 100 000 person-years (fully adjusted HR vs csDMARD, 0.93; 95% CI, 0.85-1.01) (Table 2). Correspondingly, the rate among initiators of TNFi as second bDMARD was 917 per 100 000 person-years (fully adjusted HR, 0.89; 95% CI, 0.76-1.04). Results were similar for the outcome of first invasive solid malignant neoplasm. For the outcome of first invasive squamous cell skin cancer, the HR for initiators of a first TNFi was 1.09 (95% CI, 0.84-1.42), and 0.86 (95% CI, 0.54-1.39) for initiators of TNFi as the second bDMARD. For invasive melanoma the fully adjusted HR for initiators of a first TNFi vs csDMARD was 0.84 (95% CI, 0.60-1.18). Among initiators of TNFi as the second bDMARD, the fully adjusted HR vs csDMARD was 0.94 (95% CI, 0.53-1.66). Most “on-drug” analyses yielded results similar to those of the main analysis (eTable 1 in the Supplement). Analyses for each outcome stratified by time since start of treatment (0-2, 2-4, and ≥4 years) revealed a decreased risk of overall cancer during the first 2 years of treatment for initiators of a first TNFi (Table 3). Hazard ratios vs the general population for a first invasive solid or hematologic malignant neoplasm, excluding NMSC, fell close to 1.00 (eTable 2 in the Supplement).

The rate for invasive solid or hematologic malignant neoplasms, excluding NMSC, was 959 per 100 000 person-years in the tocilizumab cohort, 1026 per 100 000 person-years in the abatacept cohort, and 1074 per 100 000 person-years in the rituximab cohort (Table 2). Fully adjusted point estimates did not reveal any statistically significant risk differences (tocilizumab vs csDMARD: HR, 0.89 [95%, CI 0.67-1.18]; abatacept vs csDMARD: HR, 0.88 [95% CI, 0.68-1.14]; and rituximab vs csDMARD: HR, 0.86 [95% CI, 0.73-1.03]).

Based on 17 events of squamous cell skin cancer in the abatacept cohort, an increased risk vs csDMARD was observed (HR, 2.15; 95% CI, 1.31-3.52). “On-drug” analyses yielded results similar to the main analysis, for all outcomes (eTable 1 in the Supplement). Hazard ratios vs the general population for a first invasive solid or hematologic malignant neoplasm excluding NMSC fell close to 1.00 (eTable 2 in the Supplement).

Apart from an increased risk for squamous cell skin cancer when comparing initiators of abatacept with initiators of a first or second TNFi, no other statistically significant risk differences for any of the outcomes for tocilizumab, abatacept, and rituximab were observed, with most HRs close to 1.00 (Table 4).

For the combined outcome of a first invasive solid or hematologic malignant neoplasm, excluding NMSC, the HR adjusted for age, sex, and start-year was 1.11 (95% CI, 1.01-1.22) (eTable 3 in the Supplement). There were also statistically significantly increased risks vs the general population for a first invasive hematologic malignant neoplasm (HR, 1.56; 95% CI, 1.13-2.16), and a first invasive squamous cell skin cancer (HR, 1.48; 95% CI, 1.03-2.13). There was a nonstatistically significant risk increase for a first invasive solid malignant neoplasm excluding NMSC (HR, 1.08; 95% CI, 0.97-1.19). On-drug analyses yielded similar results.

Alternative definitions of the bDMARD-naive RA comparator did not alter the pattern of HRs, nor did the addition of a lag period of 3 months between inclusion and start of follow-up (eTables 4 and 5 in the Supplement). eTable 2 in the Supplement presents HRs comparing all RA cohorts to the general population, adjusting for age, sex, and start year.

In this study, to our knowledge one of the largest observational studies on the risk of malignant neoplasms in RA patients treated with bDMARDs to date, we found that the overall risk of developing cancer among patients with RA initiating treatment with TNFi or non-TNFi, as used in clinical practice, did not differ from that of patients with bDMARD-naive RA.

Our current assessment represents an extension of our previous studies based on data from SRQ/ARTIS.^9,30 We found no increased risk of malignant neoplasms overall, with 95% CIs excluding clinically meaningfully increased risks. In addition, we found no increase in risk among patients starting TNFi as second bDMARD. Comparing this finding with those of our previous reports,^9,30 the difference is mainly with the patients under study; the indication for starting TNFi in RA has broadened substantially over time,⁹ such that the characteristics of the patients in our current study are quite different from patients starting TNFi in 1999. This is of importance also for the underlying cancer risks, including their potential to be modified by TNFi. With respect to site-specific malignant neoplasms, our results seemingly do not corroborate our previous finding of an increase in melanoma.⁴ The comparability across these reports is limited (differences in inclusion and exclusion, follow-up, and patient characteristics). Modifying our current analysis so that it mimicked those of our previous reports^9,30 yielded HRs close to 1.00 (eTable 6 in the Supplement). It may be that our previous finding was spurious, or that the change in characteristics of patients starting bDMARDs in recent years compared with those starting more than a decade ago (ie, lower accumulated inflammation, lower accumulated exposure to immunosuppressive drugs) has altered the potential for TNFi and other bDMARDs to exert any effect on melanoma incidence. Only continued follow-up can indicate whether there is a true calendar trend in excess risk.

For the non-TNFi, our main analysis indicated no increase in overall cancer risk with abatacept, rituximab, or tocilizumab, again with upper confidence bounds excluding clinically meaningful risks. With 1 exception (abatacept and squamous cell skin cancer), our results are thus in keeping with the limited trial data.^13-18,31,32 Although few observational studies have been published on this subject, 2 smaller studies from Finland³³ (438 patients) and France³⁴ (186 patients) did not find increased risks of malignant neoplasms among patients with RA treated with rituximab. A US cohort study³⁵ found similar risks of overall malignant neoplasm among patients with RA treated with abatacept (408 patients), or rituximab (167 patients), compared with those with RA treated with methotrexate, but, interestingly, an increased risk of NMSC among abatacept-treated patients although based on only 5 events. With a total of 25 different drug-outcome comparisons, our finding of an increased risk of invasive squamous cell skin cancer among abatacept users may be a chance finding (a Bonferroni-corrected P value for this association would be .06). Also, the point estimates for squamous cell skin cancer were attenuated when adjusting for potential confounders, although 95% CIs were wide. Nevertheless, the association may be causal; a link between squamous cell skin cancer and immune incompetence is well established. An RA retrospective study³⁶ on risk of NMSC recurrence, with data from Medicare, found a 50% increased risk of a second NMSC among patients with RA treated with combination TNFi and methotrexate therapy, compared with those receiving methotrexate monotherapy. The same study³⁶ found no statistically significant risk increases associated with abatacept and rituximab therapy, although, as the authors point out, the point estimates were similar to that of the elevated TNFi risk but the statistical precision was lower. Despite our study being the largest to date, small numbers mean that we could not ultimately rule in or out a clinically relevant risk for melanoma or for several other cancer types.

Our study has limitations. Owing to the relatively recent introduction of non-TNFi in RA treatment, follow-up time and statistical power are inherently limited to less than 10 years. Although we were able to adjust for several associated factors that could influence the choice of treatment and that might be linked to risk of developing cancer, we were probably not able to fully accommodate the potential effects of channeling to or away from specific treatments. The fact that about 80% of the patients in our study who initiated treatment with non-TNFi had previously been treated with TNFi is reflective of clinical practice in many countries and is therefore the clinically relevant group to study, but also makes it difficult to fully disentangle any effects of non-TNFi from effects of TNFi treatment, although analyses in which we adjusted for prior bDMARD use provided similar results (Table 4).

We used nationwide registers with high internal validity and coverage. Cancer outcomes were identified independently of exposure and of the treating rheumatologist. Other strengths include the new-user, active-comparator design, which decreased bias from selection and diagnostic intensity linked to DMARD intervention, such as laboratory monitoring and frequent visits. Furthermore, the choice of multiple comparator cohorts allowed for comparisons of patients faced with similar treatment scenarios, for example, the choice of initiating a second TNFi, a non-TNFi, or another csDMARD, and to evaluate whether an increased or decreased HR was specific to the comparator rather than the exposure. We were able to adjust for several potential confounders that are powerful proxies for accumulated disease severity and health care contacts.

Treatment with TNFi (as first or second bDMARD), tocilizumab, abatacept, or rituximab does not seem to increase the total occurrence of malignant neoplasms, nor did we find any signals of increased risks for specific cancer types, with 1 exception that calls for replication. While mostly reassuring, our findings do not preclude the existence of increased risks for site-specific cancers or risks that require longer time to become manifest.

Corresponding Author: Hjalmar Wadström, MD, Clinical Epidemiology Unit, Department of Medicine, Solna, Karolinska Institutet, T2:01, SE-171 76 Stockholm, Sweden (hjalmar.wadstrom@ki.se).

Accepted for Publication: July 5, 2017.

Published Online: September 18, 2017. doi:10.1001/jamainternmed.2017.4332

Author Contributions: Dr Wadström had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis

Study concept and design: All authors.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Wadström, Askling.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Wadström, Frisell.

Obtained funding: Askling.

Administrative, technical, or material support: Askling.

Study supervision: Askling.

Conflict of Interest Disclosures: The ARTIS Study Group conducts scientific analyses using data from the Swedish Biologics Register ARTIS run by the Swedish Society for Rheumatology. For these analyses, the Swedish Society for Rheumatology has received funding from MSD, BMS, Pfizer, Abbott Laboratories, SOBI, UCB, and Roche. These entities had no influence on the data collection, statistical analyses, manuscript preparation or decision to submit. They were allowed to comment on the findings prior to submission, although all final decisions resided with the investigators. Dr Askling has or has had research agreements with AbbVie, BMS, MSD, Pfizer, Roche, Astra-Zeneca, Eli Lilly, and UCB, mainly in the context of safety monitoring of biologics via ARTIS. Karolinska Institutet has received remuneration for JA participating in ad boards arranged by Pfizer and Lilly.

Funding/Support: This research was funded by the Swedish Cancer Society (Cancerfonden), the Swedish Foundation for Strategic Research (SSF), the Stockholm County Council (ALF), and the Swedish Research Council (Vetenskapsrådet).

Role of the Sponsor: The funders had no role in design and conduct of the study, the collection, management, analysis and interpretation of the data, or the preparation review, or approval of the manuscript.

Group Information: The ARTIS Study Group: Johan Askling (author), Lars Klareskog (contributor), Ronald van Vollenhoven (contributor) and Nils Feltelius (contributor) (Karolinska Institutet, Stockholm, Sweden); Eva Baecklund (contributor) (Uppsala University, Uppsala, Sweden); Alf Kastbom (contributor) (Linköping University, Linköping, Sweden); Solbritt Rantapää-Dahlqvist (contributor) and Helena Forsblad-d’Elia (contributor) (Umeå University, Umeå, Sweden); Lennart Jacobsson (contributor) (Sahlgrenska Academy, Gothenburg, Sweden); Carl Turesson (contributor) and Elisabet Lindqvist (contributor) (Lund University, Malmö and Lund, Sweden); Sofia Ernestam (contributor) (Register Holder, Swedish Rheumatology Quality Register, Sweden).