Objective To estimate (1) crude and age-and gender-adjusted incidence rates (IRs) of serious infections (SI) and (2) relative risks (RR) of SI in patients with rheumatoid arthritis (RA) initiating treatment with abatacept, rituximab or tocilizumab in routine care.
Methods This is an observational cohort study conducted in parallel in Denmark and Sweden including patients with RA in Denmark (DANBIO) and Sweden (Anti-Rheumatic Treatment in Sweden Register/Swedish Rheumatology Quality Register) who started abatacept/rituximab/tocilizumab in 2010–2015. Patients could contribute to more than one treatment course. Incident SI (hospitalisations listing infection) and potential confounders were identified through linkage to national registries. Age- and gender-adjusted IRs of SI per 100 person years and additionally adjusted RRs of SI during 0–12 and 0–24 months since start of treatment were assessed (Poisson regression). Country-specific RRs were pooled using inverse variance weighting.
Results We identified 8987 treatment courses (abatacept: 2725; rituximab: 3363; tocilizumab: 2899). At treatment start, rituximab-treated patients were older, had longer disease duration and more previous malignancies; tocilizumab-treated patients had higher C reactive protein. During 0–12 and 0–24 months of follow-up, 456 and 639 SI events were identified, respectively. The following were the age- and gender-adjusted 12-month IRs for abatacept/rituximab/tocilizumab: 7.1/8.1/6.1 for Denmark and 6.0/6.4/4.7 for Sweden. The 24-month IRs were 6.1/7.5/5.2 for Denmark and 5.6/5.8/4.3 for Sweden. Adjusted 12-month RRs for tocilizumab versus rituximab were 0.82 (0.50 to 1.36) for Denmark and 0.76 (0.57 to 1.02) for Sweden, pooled 0.78 (0.61 to 1.01); for abatacept versus rituximab 0.94 (0.55 to 1.60) for Denmark and 0.86 (0.66 to 1.13) for Sweden, pooled 0.88 (0.69 to 1.12); and for abatacept versus tocilizumab 1.15 (0.69 to 1.90) for Denmark and 1.14 (0.83 to 1.55) for Sweden, pooled 1.13 (0.91 to 1.42). The adjusted RRs for 0–24 months were similar.
Conclusion For patients starting abatacept, rituximab or tocilizumab, differences in baseline characteristics were seen. Numerical differences in IR of SI between drugs were observed. RRs seemed to vary with drug (tocilizumab < abatacept < rituximab) but should be interpreted with caution due to few events and risk of residual confounding.
- rheumatoid Arthritis
- DMARDs (biologic)
Statistics from Altmetric.com
What is already known about this subject?
Serious (=hospitalised) infections (SI) during treatment with tumour necrosis factor inhibitors (TNFi) in rheumatoid arthritis (RA) are a concern.
The risk of SI with other types of biologic disease-modifying antirheumatic drugs is less characterised.
What does this study add?
This study included >8000 treatment series of patients with RA treated with non-TNFi—abatacept (ABA), rituximab (RTX) and tocilizumab (TCZ)—in routine care in Denmark and Sweden.
Differences in baseline characteristics and numerical differences in incidence rate between the three drugs were observed.
The relative risk of SI seemed to vary modestly with drug (TCZ < ABA < RTX).
How might this impact on clinical practice or future developments?
The findings should be interpreted with caution due to the few events and the risk of residual confounding, but do not support that one non-TNFi should be recommended over the other for patients with RA at high risk of SIs.
The treatment options for patients with rheumatoid arthritis (RA) have improved significantly over the last decades.1 Biologic disease-modifying antirheumatic drugs (bDMARDs) comprise drugs with different modes of action and may be categorised as tumour necrosis factor inhibitors (TNFi) and non-TNFi bDMARDs (eg, abatacept, tocilizumab, rituximab).2 Due to their impact on the immune system, safety concerns including serious infections have been raised.3
While clinical trials generate short-term safety data in highly selected populations, observational studies based on data from clinical registries may provide information on longer term safety and safety in larger and unselected populations reflective of routine care.3–5 For patients with RA treated with TNFi, large observational studies have suggested an initial twofold increased risk of serious infections compared with biologics-naïve patients.6–8 Long-term observational studies on the risk of serious infections in patients treated with non-TNFi bDMARDs are sparse, reporting incidence rates (IRs) of serious infections between 2.8 and 18.7/100 person years (PY),9–14 which when compared with TNFi have corresponded to HRs of between 0.91 (rituximab) and 1.28 (tocilizumab).12 13 15 In all studies, CIs have been wide.12 13 15 Between-drug comparisons of the three non-TNFi bDMARDs have only been performed in one observational study, of US Medicare patients, which reported an HR of 1.36 for rituximab and 1.10 for tocilizumab compared with abatacept.14
Confounding by indication (ie, channelling of patients with certain characteristics to certain drugs) is a challenge for observational comparative safety studies, and may, if not accounted for, bias the observed safety profiles between non-TNFi bDMARDs in real-life patients.16
In light of the scarcity of comparative data on the risk of serious infections with non-TNFi bDMARDs, we took advantage of the Swedish and Danish large-scale clinical rheumatology registers and possibilities to enrich these data further through linkages to national registers. In patients with RA treated with abatacept, rituximab or tocilizumab, we aimed to estimate (1) crude as well as age- and gender-adjusted IRs of serious infections and (2) relative risks (RR) of serious infections during 0–1 and 0–2 years since treatment start, taking potential channelling into account.
Patients and methods
This observational cohort study was conducted in Denmark and Sweden in parallel. The Danish DANBIO registry covers 90%–98% of adults with rheumatic diseases treated with bDMARDs in routine care.17 18 The Swedish Anti-Rheumatic Treatment in Sweden Register (ARTIS) builds on the Swedish Rheumatology Quality Register and covers 87%–95% of bDMARD-treated patients with RA.19 Clinical data are collected prospectively in both registries.
Patients with RA were identified in DANBIO and ARTIS and included in the current study if they started treatment with abatacept, tocilizumab or rituximab between January 2010 and December 2015, either as their first ever bDMARD treatment course or with a history of treatment with another bDMARD (switchers). Patients could contribute more than one treatment course. Abatacept, rituximab and tocilizumab were made clinically available during 2008 and 2009, and therefore January 2010 was chosen as start of follow-up to allow ≥1 year of run-in of the three treatment groups.
Patient characteristics and disease activity at the date of treatment start (baseline) were retrieved from DANBIO and ARTIS. If there were no available baseline data, the first visit with clinical data from 30 days before until 6 days after baseline was used.
By use of the unique civil registration numbers, the clinical data were enriched through linkage of DANBIO and of ARTIS to each country’s National Patient Registry and National Prescription Drug Registry.20–23 The national patient registries have virtually complete data on inpatient and outpatient contacts in specialised care. When a patient is discharged from hospital, the primary and contributory diagnoses are coded by the physician (according to the International Classification of Diseases, 10th revision [ICD-10] and before 1994: ICD-8); the same applies to outpatient visits.
The main outcome was IR and RR of serious infection during treatment with a non-TNFi bDMARD. Serious infection was defined as the first hospitalisation with a primary discharge diagnosis (or contributory, if the primary diagnosis was RA) listing infection (for ICD codes, see online supplementary table S1) during follow-up from each treatment start. Serious infections were assessed from baseline until 12 months after treatment start, from baseline until 24 months after treatment start, and in consecutive 6-month intervals during the first 24 months after treatment start.
To capture infections likely attributable to the non-TNFi bDMARD, we considered patients as treated until 90 days after registered discontinuations of this treatment. Accordingly, exposure time was defined as days from baseline to the first of the following: first serious infection, 90 days after registered discontinuation of the non-TNFi bDMARD treatment, end of follow-up, emigration, start of treatment with another bDMARD, death or end of study period (Sweden: 31 December 2015; Denmark: 31 July 2015). For abatacept and tocilizumab, if the same drug was restarted within 3 months after the recorded stop, with no other bDMARD between, the treatment periods were considered as one. For rituximab, and due to the dosing regimen for this drug, the corresponding timeframe was 6 months.
The following covariates (reflecting status at start of the treatment in question) were included: (1) from DANBIO/ARTIS: calendar year, RA disease duration (years), treatment history (biologics-naïve/switch from TNFi/switch from non-TNFi bDMARD), glucocorticoid use (injections or oral) during the previous year (yes/no), Disease Activity Score of 28 joints (DAS28; remission/low/moderate/high disease activity/missing), functional status (Health Assessment Questionnaire [HAQ; continuous]), C reactive protein (CRP; ≤10 mg/L or >10 mg/L), use of any concomitant conventional synthetic disease-modifying antirheumatic drugs (yes/no), use of concomitant methotrexate (yes/no), IgM rheumatoid factor status (positive/negative), and current smoking (yes/no/missing); (2) f rom national patient registries : diagnosis (inpatient or outpatient contact) of cancer ever (yes/no), and of any of the following during the 5 years prior to treatment start (yes/no): serious infection, knee or hip prosthesis, chronic obstructive or interstitial pulmonary disease (COPD), diabetes, myocardial infarction or chronic kidney disease (see online supplementary table S1 for ICD codes), and days of hospitalisation in the year prior to treatment start (tertiles); and (3) from prescription drug registries (the year prior to baseline): number of unique (reimbursed in Denmark and dispensed in Sweden) prescription episodes (tertiles). If a patient had more than one prescription, every prescription was counted (summed up for antibiotics Anatomical Therapeutic Chemical Classification (ATC) code J01, antimycotic J04 and antiviral drugs J05).
All analyses were conducted separately in the two countries (SAS V.9.4) according to a predefined statistical analysis plan. Descriptive results are presented as median (IQR) or as percentages.
Patients who, during the study period, started more than one non-TNFi bDMARD contributed PY (and potential events) to both treatments according to the definition of time on non-TNFi bDMARD (exposure time). Thus, a patient could contribute more than one treatment course, which was accounted for in the statistical analysis with robust SEs. Crude IRs of serious infections are presented as events/100 PY with 95% CI. Crude IRs for each drug were adjusted for age and gender using Poisson regression. In sensitivity analyses, we additionally classified patients as biologics-naïve or switchers. By multivariable Poisson regression models, we conducted pairwise comparisons of the non-TNFi bDMARD to estimate RR. Potential confounders were selected among the above-mentioned covariates (adjusted for age and gender) in two ways: (1) identified a priori as clinically relevant: age, gender, DAS28, disease duration, HAQ, smoking, previous malignancy, previous serious infection, previous number of prescriptions and previous COPD; and (2) variables which in age- and gender-adjusted analyses changed the estimate markedly. Data availability is shown in online supplementary table S2.
Three multivariable models were developed, where the above-mentioned variables were included. Model A was adjusted for age and gender. Model B was adjusted for the covariates in model A plus variables from DANBIO/ARTIS (DAS28, smoking, disease duration and HAQ). Model C was adjusted for the covariates in model B plus variables from national patient registries (previous serious infection, previous malignancy, previous COPD and number of prescriptions). Finally, the RRs for both countries were pooled using inverse variance weighting to provide an overall estimate.
For the main analyses, no imputation of missing data was performed. AFor sensitivity analyses, multiple imputation by chained equation for missing data were performed.
A total of 8987 non-TNFi bDMARD treatment courses (6648 unique patients) were identified (abatacept: 2725; rituximab: 3363; tocilizumab: 2899) (table 1). In 25% of the treatment courses in Denmark, patients were biologic-naïve, in Sweden 21% were biologic-naïve. Four hundred and twenty-eight patients in Denmark and 1066 patients in Sweden contributed follow-up time to more than one treatment course. In 55% of the treatment courses, patients had previously been treated with a TNFi, and in 22% patients had previously received another non-TNFi bDMARD (both termed switchers). In both countries, use of rituximab was associated with higher age, longer disease duration and more often with a history of malignancy. Tocilizumab was associated with higher baseline CRP (table 1). In biologics-naïve patients who started abatacept (n=601), rituximab (n=958) or tocilizumab (n=461), a similar pattern for rituximab was observed (online supplementary table S3).
IRs of serious infections
During the first 12 months from treatment start, 456 serious infections were identified (abatacept: 130; rituximab: 206; tocilizumab: 120) (table 2), of which 299 (66%) occurred within the first 6 months of treatment (table 3). The corresponding numbers during the first 24 months were 639 (abatacept: 169; rituximab: 310; tocilizumab: 160) (table 2). In the first 12 months after treatment start, the highest age- and gender-adjusted IR of infection was observed in patients treated with rituximab, followed by abatacept and tocilizumab, in both Denmark and Sweden (table 2). However, the 95% CIs were wide and overlapping. The Danish rates were consistently higher than the Swedish. When stratifying the analysis by biologics-naïve patients and switchers, a similar pattern was found (online supplementary table S4). Similar results were observed when follow-up was extended from 12 to 24 months (table 2).
Analyses by 6-month intervals showed that age- and gender-adjusted IRs of serious infections were highest during the first 6 months for patients treated with rituximab (Denmark: 11.0 [7.8 to 15.6]/100 PY; Sweden: 7.5 [5.9 to 9.5]/100 PY) and tocilizumab (Denmark: 7.1 [5.0 to 10.0]/100 PY; Sweden: 5.5 [4.0 to 7.4]/100 PY) and patients treated with abatacept in Denmark (9.7 [6.8 to 13.8]/100 PY) (figure 1,table 3). For Swedish patients treated with abatacept, the age-adjusted and gender-adjusted IRs were similar from 0 to 6 and 6–12 months (5.9/100 PY) (figure 1, table 3). The IRs during the first 6 months stratified by biologics-naïve and switchers showed the same pattern (online supplementary table S4).
Country-specific RR of serious infections across the three non-TNFi bDMARDs
Differences in RRs across the three non-TNFi bDMARDs were observed. During the first 12 months since treatment start, the RR for serious infection (model A) for tocilizumab versus rituximab was 0.76 (0.52 to 1.11) in Denmark and 0.75 (0.56 to 1.00) in Sweden, 0.89 (0.58 to 1.34) in Denmark and 0.94 (0.72 to 1.22) in Sweden for abatacept versus rituximab, and 1.17 (0.77 to 1.76) in Denmark and 1.26 (0.93 to 1.71) in Sweden for abatacept versus tocilizumab (figure 2, table 4). Thus, the directions were the same in Denmark and Sweden, but all 95% CIs were wide and included 1. Overall, further adjustment (eg, for treatment history, glucocorticoids, disease duration and comorbidities) in models B and C only had minor impact on the RR estimates (table 4 and online supplementary table S5). Similar estimates were found after 24 months (table 4, online supplementary table 6 and online supplementary material 1). Multiple imputation of missing data did not substantially change the estimates (online supplementary table S5).
Pooled RR of serious infections across the three non-TNFI bDMARDs
The RRs from model C during the first 12 months since treatment start for Denmark and Sweden combined were 1.13 (0.91 to 1.42) for abatacept versus tocilizumab, 0.88 (0.69 to 1.12) for abatacept versus rituximab, and 0.78 (0.61 to 1.01) for tocilizumab versus rituximab. All 95% CIs for model C included 1 (figure 2, table 4). Similar RRs were found for models A and B, but the CIs for the comparison between tocilizumab and rituximab did not include 1, that is, reached statistical significance (table 4).
In this observational study of more than 8000 non-TNFi bDMARD treatment courses, we investigated the risk of serious (hospitalised) infection in patients with RA treated with abatacept, rituximab or tocilizumab in routine care in Denmark and Sweden. The rate of infections ranged from 4.7 to 8.1/100 PY during the first year and from 4.3 to 7.5/100 PY during the second year after treatment start. The patterns of risks and relative risks across treatments were consistent for Denmark and Sweden, with a ≈25% lower observed RR of infection during the first year after treatment start for tocilizumab and ≈10% lower RR for abatacept compared with rituximab, and ≈20% higher RR for abatacept compared with tocilizumab. However, the observed differences in risk should be interpreted with caution, considering potential channelling caused by confounding by indication and residual confounding.
As expected, since treatment with these drugs was not randomised, differences in baseline characteristics between the three patient groups were observed in both countries. Channelling of, for example, older patients with more comorbidities to certain treatments increases the risk of a serious infection, as shown in previous studies of mainly TNFi-treated patients with RA.24–27 To accommodate such confounding by indication, we adjusted for treatment history, previous infections and comorbidities as measures of general frailty. However, this generally had limited impact on the observed age- and gender-adjusted RR. Few events in some patient strata may have affected the model.
The risks of serious infections were highest during the first 6 months after treatment start for all three drugs. This was most prominent for Danish patients treated with abatacept and rituximab, where a steep decline in IR after 6 months was observed (and based on small numbers). This decrease in risk corresponds to what has previously been reported from observational studies on TNFi6 28 and may represent a ‘healthy user effect’29 with a time-dependent depletion of patients susceptible to serious infections. The same tendency was observed when the analysis was stratified by biologics-naïve and switchers.
Observational studies on the risk of serious infections in TNFi-treated patients with RA have reported IRs between 1.66 and 5.4 per 100 PY.6 28 30 These rates appear slightly lower than those observed in our study. This might reflect that patients who start treatment with non-TNFi bDMARDs represent a more selected cohort of patients. Thus, in most of the treatment courses, the patients in the present study had previously failed another bDMARD (switchers) or had previous malignancy or other risk factors that contraindicated treatment with TNFi. A meta-analysis on the risk of serious infections for non-TNFi bDMARDs based on data from randomised controlled trials (RCTs) and long-term extension studies reported lower IRs than in our study: 3.04 for abatacept, 3.72 for rituximab and 5.45 per 100 PY for tocilizumab.31 This difference may be explained by the strict inclusion criteria for patients participating in RCTs. Only few observational studies of the risk of serious infections during treatment with non-TNFi bDMARDs in RA have been published, and they have reported IRs between 2.8 and 18.7.9 11–15 In patients treated in routine care, comparison of risk between non-TNFi bDMARDs have mainly been indirect with TNFi as reference drug, and the studies have shown partly conflicting results.11–15 One study compared tocilizumab with TNFi treatment (HR=1.28).13 Another compared rituximab and tocilizumab with etanercept and reported, in contrast to our study, a lower risk for rituximab (HR=0.91) and higher for tocilizumab (HR=1.21).15 A third study compared abatacept with rituximab and found higher risk in abatacept-treated patients (HR=1.21).12 In all studies, the CIs were wide. An American study compared the three non-TNFi bDMARDs.14 With abatacept as reference, they reported an HR of 1.36 for rituximab and 1.10 for tocilizumab, the latter conflicting our findings. In the same study, crude IRs for abatacept, tocilizumab and rituximab of 13.1/100 PY, 14.9/100PY and 18.7/100PY, respectively, were found during the first year of treatment.14 The study included 31 801 new biologic treatment episodes in patients who had previously received another biologic agent and was based on a national insurance programme with mainly elderly (≥65 years) or disabled individuals. This might explain the considerably higher rates and make results less generalisable. A study from the German biologic registry RABBIT (Rheumatoid Arthritis: Observation of Biologic Therapy), reporting infection risk versus TNFi, showed a 15% higher infection risk for tocilizumab, an 8% lower risk for rituximab and an 18% lower risk for abatacept.32 The estimates arising from these indirect comparisons are not in agreement with our direct comparisons, where rituximab had the highest risk, followed by abatacept and tocilizumab.
Concomitant treatment with glucocorticoids is known to increase risk of serious infections.33 34 High-dose glucocorticoids are routinely given prior to rituximab infusions. Furthermore, glucocorticoids are used as bridging therapy. We adjusted RRs for the use of glucocorticoids (oral and parenteral), obtaining information from DANBIO and ARTIS, but it did not change the estimates substantially. Some missing/incorrect registration may occur in these data from the registries.
This study has strengths and limitations. High-quality clinical prospective data from nationwide rheumatological registries were enriched with data from virtually complete national registries. Including data from two countries increased the number of events and the external validity. We observed similar interdrug relationships infection risk for both countries, suggesting that our results are not explained by country-specific factors. Loss to follow-up was negligible, thus reducing the risk of selection bias. Limitations included the risk of residual confounding due to the observational design. Also, information bias may have been present since some missingness occurred. This was addressed using multiple imputation analysis. There may be some misclassification of exposure; for example, patients who go into remission withdraw from treatment without informing the hospital. Furthermore, misclassification of outcomes cannot be ruled out, for example, that fragile patients are hospitalised for less severe infections than younger patients are.
In conclusion, this collaborative project between Denmark and Sweden including more than 8000 treatment courses of patients with RA treated with abatacept, rituximab and tocilizumab in routine care demonstrated differences in baseline characteristics of the treated patients. IRs declined over time and numerical differences between the three drugs were observed. The RR of serious infection seemed to vary with drug (tocilizumab < abatacept < rituximab) but should be interpreted with caution due to the few events and the risk of residual confounding.
Thanks to Laura Köcher for help with the figures, and to the Danish and Swedish departments of rheumatology that contributed data.
Handling editor Josef S Smolen
Collaborators The ARTIS Study Group: Johan Askling (author), Lars Klareskog (contributor) and Nils Feltelius (contributor) (Karolinska Institutet, Stockholm, Sweden); Eva Baecklund (contributor) (Uppsala University, Uppsala, Sweden); Christopher Sjöwall (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 (author) and Elisabet Lindqvist (contributor) (Lund University, Malmö and Lund, Sweden); Ralph Nisell (contributor) (Register Holder, Swedish Rheumatology Quality Register, Sweden).
Contributors KLG, BG, MLH, MN, MØ, LD, EVA and JA contributed to the study design. KLG, BG, MLH, NSK, FM, EVA and JA contributed to data management. All authors contributed to analyses of raw data and interpretation. All authors contributed to and approved the final manuscript.
Funding The study is partly funded by NordForsk and Foreum.
Competing interests KLG: BMS. EVA: none. BG: Biogen, AbbVie and Pfizer. FM: none. MØ: AbbVie, BMS, Boehringer Ingelheim, Celgene, Eli Lilly, Hospira, Janssen, Merck, Novartis, Orion, Pfizer, Regeneron, Roche and UCB. LD: UCB, MSD, Eli Lilly and Janssen Pharmaceuticals. MN: none. NSK: none. MLH: AbbVie, Biogen, BMS, Celltrion, MSD, Novartis, Orion, Pfizer, Samsung and UCB. JA: AbbVie, BMS, MSD, Pfizer, Roche, AstraZeneca, Eli Lilly, Samsung Bioepis and UCB, mainly in the context of safety monitoring of biologics via ARTIS. Karolinska Institutet has received remuneration for JA participating in advisory boards arranged by Pfizer and Eli Lilly.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Author note The DANBIO registry has entered into agreements with AbbVie, Biogen, BMS, Eli Lilly, MSD, Novartis, Pfizer, Roche and UCB. They receive postmarketing data and had no influence on the data collection, statistical analyses, manuscript preparation or decision to submit. The ARTIS Study Group conducts scientific analyses using data from the Swedish biologics register ARTIS, run by the Swedish Society for Rheumatology. ARTIS has entered into agreements with AbbVie, BMS, Lilly, Merck, Pfizer, Roche, Samsung Bioepis and UCB. These entities had no influence on the data collection, statistical analyses, manuscript preparation or decision to submit. BMS and Roche were allowed to comment upon the findings prior to submission, although all final decisions resided with the investigators.
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.