Rheumatoid arthritis (RA) is a well-established risk factor for malignant lymphomas. Cohort studies typically suggest a doubling of the occurrence of lymphoma.1^–8 Recent studies have indicated a strong association between RA disease intensity and lymphoma risk,7 9 but less evidence of any direct link between disease-modifying antirheumatic drugs (DMARD)9 10 or steroids9 11 and the risk of lymphoma.

Ever since the first human studies on tumour necrosis factor (TNF) antagonists, their safety profile with respect to the risk of malignant lymphomas has been a concern,12 as reflected by their product labels.13^–15 Although no individual clinical trial has so far reported a statistically increased occurrence of lymphoma among patients exposed to TNF antagonists, none has been adequately powered to address this issue. A recent grouped-data meta-analysis of controlled RA trials of infliximab and adalimumab indicated a conspicuously increased short-term (less than one year) occurrence of malignant lymphomas during and after the trial phase.16 Some observational studies, of highly different design, setting and case mix, of patients exposed to TNF antagonists in clinical practice have suggested the possibility of an increased risk,17 but others have failed to confirm this,8 18 as was the case for an interim analysis from our group.5 Several of these studies have, however, reported relative risks (RR) slightly above rather than below unity,5 18 and in each study the low number of lymphoma cases has prevented the exclusion of a clinically meaningful increase in risk following TNF antagonist treatment. Beyond the reported overall estimates of RR, there is a scarcity of data on potential heterogeneities in this risk, eg, in different treatment settings, by time since the start of anti-TNF treatment and by the accumulated duration of treatment.

In this study, which represents a more than doubled number of patients, time of follow-up and observed lymphomas compared with our previous report,5 we aimed at a better estimation of the overall association between anti-TNF therapy and lymphoma risk. In addition, we aimed to assess the potential heterogeneities in this risk with respect to patient selection and calendar time. To do this, we took advantage of the unique opportunities offered by the Swedish population-based healthcare, mandatory cancer reporting, the nationwide Swedish Biologics Register and linkage with other national health and population registers.

SUBJECTS AND METHODS

Setting

The public, population and non-insurance-based Swedish healthcare system, the RA cohorts, the national registers and the personal identification system allowing for record linkage have all been described elsewhere.19

Study population

A national cohort of Swedish patients with RA

By linking together data from two pre-existing and overlapping data sources including information on patients with RA, we assembled an RA cohort that encompassed the vast majority of all RA patients in this country alive in 1998, when anti-TNF therapy was first available in Sweden (and hence the start of our study period). In the nationwide and population-based Swedish Inpatient Register, which contains information on all inpatient care (dates, medical discharge diagnoses, coded according to International Classification of Diseases versions 7–10),20 department, hospital, personal identification number since 1964, (nationwide since 1987), we identified all 51 588 individuals hospitalised with (but not necessarily because of) RA between 1964 and 2005. Previous validation surveys against information in the underlying medical files suggest that the diagnostic correctness of these RA diagnoses is approximately 90%.9 In the same register, we also identified all 45 903 patients who had a registered outpatients (non-GP) visit listing RA in 2001–5. In the Swedish Early RA Register, into which incident RA has been registered since the mid-1990s, we identified 5529 incident cases with RA diagnosed between 1995 and 2006. There was a considerable overlap between the three above RA sources, such that the total number of RA patients in the assembled RA comparator was 67 743 (73% women, mean year of birth 1936, mean age at start of follow-up 64 years). Defining RA through this cohort, the estimated population prevalence in Sweden in 2005 would be approximately 0.5%, well in line with previous estimates.21

Anti-TNF therapy

Since 1998, patients above 16 years of age with RA starting anti-TNF therapy have been entered and followed up in the practice-based national Swedish Biologics Register (ARTIS).19 For each initiated treatment, information on the underlying rheumatological condition including date of onset, date of treatment initiation (and discontinuation), type and dose of biological agent, disease activity score based on 28 joints (DAS28) and health assessment questionnaire (HAQ), concomitant DMARD, steroids, non-steroidal anti-inflammatory drugs and analgesics are recorded by the treating rheumatologist at treatment start as well as at prespecified follow-up visits. Apart from baseline and follow-up-specific data on RA activity and treatment, ARTIS also includes a reporting system for adverse events.

By linking the above-mentioned national cohort of 67 743 patients with RA to the Swedish Biologics Register, we identified those individuals starting on anti-TNF therapy, including their dates of start/stop. A total of 6604 patients with RA started a first TNF antagonist treatment between 1998 and 31 July 2006 (table 1); 51% were first treated with infliximab, 34% with etanercept and 14% with adalimumab; 25% had been treated with more than one anti-TNF drug. The distribution of patient characteristics (table 1) was strikingly similar for etanercept, infliximab and adalimumab apart from a higher proportion of concomitant methotrexate use among patients starting infliximab (not shown). There was, however, a linear trend with declining severity at first treatment start from 1999 to 2006, both with respect to measures of disease activity at the time of treatment start (eg, DAS28) and with respect to measures of accrued disease severity up until the time of treatment start (eg, large joint replacements, accrued number of days spent in hospital) although the sex and age distributions, and the occurrence of other specific co-morbidities, were similar (table 1).

General population comparator cohort

For each unique individual with RA, randomly selected individually matched (sex, year of birth, county of residence, civil status) general population controls were identified in the Swedish Population Register. These controls were conditioned to be alive at the first entry 1964–2005 of their corresponding RA patient into our RA cohort. In total, 471 024 general population controls alive at the earliest start of our study period (1 January 1998) were selected.

Lymphoma occurrence

Linkage to the Swedish Cancer Register, to which reporting of all malignant neoplasms is mandatory both for the treating physician and for the pathologist since 1958,22 identified all registered malignant lymphomas and other cancers between 1958 and December 2006 for all unique individuals in the study. Previous reclassification of some 400 RA lymphomas identified through this method indicates a diagnostic correctness greater than 95% with respect to the lymphoma diagnosis.9 For RA patients exposed to TNF antagonists, we also searched for lymphomas among the adverse events reported within the framework of the Swedish Biologics Register (no additional cases were detected), and verified all lymphomas through scrutiny of the medical records and reclassification of the tissue specimens.

Vital status during follow-up

Through their personal identification numbers, all individuals were linked to the Swedish Cause of Death Register (date of death) and to the Swedish Population and Emigrations Register (residency and emigrations during the study period). Some 0.3% of all individuals were excluded because of data inconsistencies precluding unambiguous follow-up throughout our study period (1998 to 31 December 2006).

Statistical analyses

Incidence of malignant lymphomas after start of first TNF antagonist treatment

The incidence of lymphomas in RA patients exposed to anti-TNF therapy was calculated counting the time from first TNF antagonist start until the first of: lymphoma diagnosis, death, emigration, or end of the study period, overall and stratified by sex, age at follow-up, time since treatment start, accumulated time on active anti-TNF therapy, and calendar period of start of first anti-TNF treatment.

Relative risk of lymphomas in RA treated with anti-TNF therapy versus anti-TNF-naive RA patients

RR comparing RA patients in the national RA cohort treated or not with anti-TNF therapy was assessed through Cox regression using anti-TNF therapy as a time-dependent variable, so that patients were considered unexposed until first treatment start and as exposed thereafter. Cox models used calendar time as the time-scale and were stratified for sex, civil status, county of residence and year of birth in one-year strata such that RR were assessed controlling for attained age and calendar time. In the assessment of drug-specific RR patients were considered as exposed to their first anti-TNF drug until the date of any first switch to a second anti-TNF drug and as switchers thereafter. All analyses were adjusted for co-morbidities and the use of inpatient care until 6 months before lymphoma occurrence (hip, knee, shoulder and ankle-joint replacement, diabetes, cardiovascular disease, infection, chronic obstructive pulmonary disease, accumulated number of hospitalisations, accumulated number of RA hospitalisations and accumulated number of days spent in hospital), as well as their source cohort, all treated as time-dependent covariates.

Relative risk of lymphomas in RA patients treated with TNF antagonists versus the general population

RR for the comparison of RA patients treated with anti-TNF therapy versus the general population comparator were assessed using Cox regression stratified by sex, year of birth, civil status and county of residence, using calendar time as the time scale.

The study was approved by the Ethics Committee at Karolinska Institutet.

RESULTS

Incidence of malignant lymphomas after start of first TNF antagonist treatment

Among the 6604 RA patients treated with TNF antagonists, 26 malignant lymphomas occurred during 26 981 person-years of follow-up since the start of the first TNF antagonist treatment, corresponding to an overall lymphoma incidence of 96 (95% CI 63 to 141) per 100 000 (table 2).

As expected from the occurrence of lymphomas in the general population, the incidence increased with age and was higher among men than women. Whereas the lymphoma incidence was stable across time intervals since the start of anti-TNF therapy and the accumulated duration on active anti-TNF therapy, respectively, the incidence dropped by a factor three between RA patients starting anti-TNF therapy in 1998–2001 (122 per 100 000) and those starting similar treatment in 2004–6 (43 per 100 000, table 2). When the lymphoma incidence by the calendar period of the first start of anti-TNF therapy was stratified by the time since the start of follow-up the tendency towards higher incidences among early starters (1998–2001) did not appear to be due to the fact that they had longer follow-up times (table 3).

Compared with all other patients starting on anti-TNF therapy, the 26 patients developing lymphoma were characterised by somewhat higher disease activity at the time of treatment start and by somewhat more accrued hospitalisations, although not markedly different from those RA patients starting anti-TNF therapy in 1998–2001 (19 out of the 26 lymphomas started anti-TNF therapy in 1998–2001, table 1 and table 2).

Relative risk of lymphomas in RA patients treated with anti-TNF therapy versus anti-TNF-naive RA patients

Compared with the 336 lymphomas that occurred during 365 026 person-years of anti-TNF-naive follow-up in the national RA cohort of 67 743 patients, the 26 lymphomas occurring against a background of anti-TNF therapy were associated with a statistically non-significant increase in lymphoma risk, adjusted RR of 1.35 (95% CI 0.85 to 2.11, unadjusted RR 1.30), which was entirely accounted for by those patients starting anti-TNF therapy in 1998–2001 (RR 1.62) rather than those starting in 2002–6 (RR 0.90, table 2). There was no trend in RR with increasing time since start of anti-TNF therapy, nor with accumulated time on active anti-TNF therapy (table 2). We did not note any major difference in RR for each of the three anti-TNF drugs (RR for etanercept, infliximab, adalimumab and for switchers all fell between 1.2 and 1.5, not shown) although precision was limited.

Relative risk of lymphomas in RA patients treated with TNF antagonists versus the general population

In the general population comparator, 1568 lymphomas occurred among 471 024 subjects during 3 355 849 person-years of follow-up from 1998 to 2006. Compared with this, anti-TNF therapy was associated with a tripled lymphoma risk (RR 2.72, 95% CI 1.82 to 4.08), which did not vary appreciably with sex or age at treatment start, nor with the time since first treatment start, but was higher among RA patients first starting anti-TNF therapy in 1998–2001 (RR 3.50) than among those starting TNF antagonist treatment in 2004–6 (RR 1.36), table 2.

DISCUSSION

In this national and population-based assessment of lymphoma occurrence among Swedish RA patients treated with anti-TNF therapy in clinical practice, RA patients treated with anti-TNF agents were at a clearly increased lymphoma risk compared with the general population and at a slight but non-significant increased risk also compared with RA patients not treated with anti-TNF therapy. Importantly, however, stratified analyses suggested that these increases were accounted for by those patients who were the first to receive anti-TNF therapy. No increase was observed among RA patients first starting anti-TNF therapy in 2002 or later. By contrast, we noted no increase in lymphoma risk shortly after the start of anti-TNF therapy, nor any increase in lymphoma risk with increasing time since treatment start, accumulated time on active anti-TNF therapy, or by any particular anti-TNF drug.

Published data on lymphoma risk associated with anti-TNF therapy have presented inconclusive and to some extent contradictory results. In several of the anti-TNF RA trials,13^–15 and in the grouped-data meta-analysis by Bongartz et al16 (later supplemented with additional trials),23 a higher occurrence of malignancies including lymphomas was observed in trial arms treated with infliximab or adalimumab than in the comparator arms. In that meta-analysis, the elevated risk of lymphomas was based on four lymphomas in 3439 patients (incidence assuming an average 6-month trial phase of 230 per 100 000) compared with none out of 1512 patients during the trial phases, which were 54 weeks or less. Moreover, a conspicuous number of lymphomas was reported after the trial phase.16 Using non-population-based US/Canadian claims data of RA patients above 65 years of age, Setoguchi et al18 assessed the lymphoma risk associated with the use of biological agents. Based on four lymphomas in 1512 biologicals users (incidence 135 per 100 000) compared with 54 lymphomas among 6318 non-users, the crude RR was 1.20 (1.11, 95% CI 0.57 to 2.51) following adjustments. Risks stratified by time or drug were not presented. In a first study, Wolfe and Michaud24 used data from the National Databank for Arthritis and observed a moderately higher risk for biologicals-exposed RA patients than for biologicals-naive RA patients, based on 14 exposed cases (incidence 140 per 100 000, RR 1.3). A recent extension encompassing some 10 000 patients exposed to anti-TNF therapy reported an overall RR of 1.0 (0.6 to 1.8).8 Data from the Swedish Biologics Register has been used in two previous reports. In a study by some of us, a geographically defined subset of the Swedish Biologics Register 1999–2002 was compared with respect to lymphoma incidence and through linkage to a cancer register to an RA community cohort from the same region. Based on five cases among 757 biologicals users (incidence 311 per 100 000) compared with two cases among 800 non-biologicals users, a fivefold but statistically imprecise increase in lymphoma risk was observed.17 In a subsequent national assessment during the study period 1998–2003, we noted nine lymphomas (incidence 93 per 100 000) and a RR of 1.1 (95% CI 0.6 to 2.1) compared with an inpatient RA comparator.5

In comparison with the above, our current report represents a considerable extension, allows for some subset analyses and is one of the largest and few population-based risk assessments of anti-TNF therapy so far. In comparison with previous studies based on patient or doctor-reported events,8 24 one particular strength of our assessment is the use of an external, independent and virtually complete ascertainment of lymphoma occurrence (the Swedish Cancer Register) and vital status during follow-up. This outcome ascertainment minimised bias from selective recall or reporting related to knowledge of treatment status.

Some limitations of our study should be pointed out. In ARTIS, patients who started TNF antagonist treatment when these drugs first became available had a higher DAS28/HAQ at treatment start, a more extensive history of the use of inpatient care and a higher incidence of lymphoma than those starting the first TNF antagonist treatment during the past 5 years. Considering the natural course of the implementation of new drugs in clinical practice, these contrasts are anticipated. The early (1998–2001) starters of TNF antagonists also accounted for the clear (vs the general population comparator) and potential (vs the RA comparator) increase in lymphoma risk. Given the known association between RA severity and lymphoma risk,9 it may be that the difference in observed RR among early and later starters of anti-TNF therapy (vs both comparators) reflects a limited ability of our study to adjust fully for accrued disease severity and that better means to quantify and adjust for pretreatment lymphoma risk would have reduced this difference. Importantly, even if so, the conclusion that channelling bias rather than intrinsic drug effects make up the increased risk would remain. It cannot be excluded, however, that under some conditions that applied to early (1998–2001) but not later (2002–6) starters TNF antagonism may act to unleash incipient lymphomas. We note, however, that the lymphoma incidence among early starters in our study was elevated throughout follow-up rather than being confined to the first months after the start of treatment. The comparatively low and declining incidence of lymphoma among those starting anti-TNF therapy in 2002–6 might suggest a beneficial effect of anti-TNF therapy on lymphoma occurrence in RA. Presently, however, it would be premature to rule out any increase in lymphoma risk with anti-TNF therapy, as well as to conclude that prolonged therapy definitely reduces the elevated risk of lymphoma. Also, a number of alternative explanations for the difference in RR between early and late starters should be considered: (1) a chance finding; the number of lymphoma cases in the anti-TNF cohort was too small to allow for temporal trends to become significant and (2) a critical dependence on the definition of “early” and “later”; although the use of alternative categorisations, eg, 1998–2002 and 2003–6, rendered very similar results.

In conclusion, our study suggests that anti-TNF therapy per se does not markedly increase the risk of lymphoma in contemporary patients with RA, but that the inherent lymphoma risk in RA patients starting anti-TNF therapy may have changed over time. In demonstrating a calendar trend in lymphoma incidence, our study also points to the need for protracted follow-up of the safety profile of biological agents and the limitations related to inferences based on overall single-point estimates of risk.