To determine the short-term and medium-term risks of cancer in patients receiving anti–tumor necrosis factor α (anti-TNFα) therapies that have proven effective in the treatment of chronic inflammatory conditions.
To determine the short-term and medium-term risks of cancer in patients receiving anti–tumor necrosis factor α (anti-TNFα) therapies that have proven effective in the treatment of chronic inflammatory conditions.
By linking together data from the Swedish Biologics Register, Swedish registers of RA, and the Swedish Cancer Register, we identified and analyzed for cancer occurrence a national cohort of 6,366 patients with RA who first started anti-TNF therapy between January 1999 and July 2006. As comparators, we used a national biologics-naive RA cohort (n = 61,160), a cohort of RA patients newly starting methotrexate (n = 5,989), a cohort of RA patients newly starting disease-modifying antirheumatic drug combination therapy (n = 1,838), and the general population of Sweden. Relative risks (RRs) were estimated using Cox regression analyses, examining overall RR as well as RR by time since the first start of anti-TNF therapy, by the duration of active anti-TNF therapy, and by the anti-TNF agent received.
During 25,693 person-years of followup in 6,366 patients newly starting anti-TNF, 240 first cancers occurred, yielding an RR of 1.00 (95% confidence interval 0.86–1.15) versus the biologics-naive RA cohort, and similar RRs versus the other 2 RA comparators. RRs did not increase with increasing time since the start of anti-TNF therapy, nor with the cumulative duration of active anti-TNF therapy. During the first year following the first treatment start, but not thereafter, dissimilar cancer risks for adalimumab, etanercept, and infliximab were observed.
During the first 6 years after the start of anti-TNF therapy in routine care, no overall elevation of cancer risk and no increase with followup time were observed.
Tumor necrosis factor α (TNFα) therapy exerts biologic effects on carcinogenesis and tumor progression, the impact of which is incompletely understood but is presumably different for different types of cancer and for different time points during carcinogenesis (1–3). Accordingly, the consequences of anti-TNF treatment on the short-term and long-term occurrences of cancer remain a concern (4).
With respect to short-term cancer risk, meta- analyses of clinical trial data have indicated the possibility of a markedly increased risk of several types of cancer occurring within months of the initiation of treatment (5–7), as have the results from at least 2 of the ∼75 individual controlled trials hitherto performed (8, 9). So far, observational data, including an interim analysis conducted by our group (10), have typically not been able to replicate this increase in short-term overall risk (11, 12), although the numbers have been small to moderate. It has been suggested that 1 reason for the discrepancy between data from trials and from observational studies lies in the choice of comparator in the observational studies. Specifically, screening and evaluation before starting/changing any RA treatment might inadvertently, but selectively, deplete the treated cohort of its members with incipient cancers. Accordingly, RA patients newly starting therapy with anti-TNF should be compared with RA patients newly starting therapy with other agents for this disease (13).
Considering that the typical period of time from the initiation of a cancer, through the development of dysplasia, and until the clinical manifestations of the cancer are seen is counted in years rather than in months and that the lifetime risk of cancer in humans exceeds 20% (14), any long-term effects of therapy are not only of high importance, but they are also impossible to address using data from clinical trials. So far, only 1 (non–population-based) observational study has reported on cancer risks as a function of time since treatment start (12), and no study has assessed cancer risks in relation to the duration of active therapy.
In this study, which represents one of the largest and longest population-based assessments of cancer risks associated with anti-TNF therapy, we therefore assessed the risk of cancer in a national cohort of Swedish RA patients treated with anti-TNF therapy. In particular, we assessed risks in relation to the time since starting the first anti-TNF agent (irrespective of treatment discontinuation), the duration of anti-TNF therapy, and the specific anti-TNF agent received.
The public, population-based, and non–insurance-based Swedish health care system, the RA cohorts, the population-based mandatory and virtually complete national registers, and the personal identification system allowing for record linkage have all been described elsewhere (15). This study was approved by the Ethics Committee at Karolinska Institute.
A national cohort comprising the vast majority of all patients with RA who were alive in 1998 (when anti-TNF was first available in Sweden; hence, the start of our study period) was assembled from 3 preexisting and overlapping data sources: the Swedish Inpatient Register, the Swedish Outpatient Register, and the Swedish Early RA Register.
From 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 (16)], admitting department, hospital, personal identification number) since 1964 (nationwide since 1987), we identified all 51,588 individuals hospitalized with, but not necessarily because of, RA between 1964 and 2005. Previous validation surveys against information in the underlying medical files suggested that the diagnostic correctness of these RA diagnoses is ∼90% (17).
From the Swedish Outpatients Register, which contains >90% of all non–general practitioner (non-GP) visits in Sweden since 2001, we identified all 45,903 patients who had a registered outpatient visit to a non-GP that listed RA as a diagnosis between 2001 and 2005. Validation surveys against information in the underlying medical files suggested that the diagnostic correctness of these RA diagnoses is also ∼90% (Askling J: unpublished observations).
From the Swedish Early RA Register, into which incident RA has been registered since the mid-1990s, we identified 8,084 incident cases of RA diagnosed from 1995 through 2007. Since the average accrued followup time since inclusion in the Early RA Register is currently >5 years, this cohort should be viewed as an “incident RA” cohort rather than specifically an “early RA” cohort.
There was considerable overlap between these 3 RA sources, such that the total number of RA patients who were alive at the beginning of our study period in 1998 and contributed to the RA cohort was 67,743 (73% women, with 1936 as the mean year of birth with a mean age of 64 years at the start of followup). Defining RA in the general population through this cohort, the estimated prevalence of RA in Sweden in 2005 would be ∼0.5%, which is consistent with previous estimates (18), suggesting that the cohort comprised the vast majority of all RA in the study base. Assessments of cancer incidence in this national RA cohort has indicated a 10–15% overall increase in the risk of cancer compared with the general population, which is consistent with that in previous studies (12, 19).
Since 1998, patients with RA (or other rheumatologic diseases) who are older than 16 years and are starting treatment with TNF antagonists have been entered into, and followed up in, Anti-Rheumatic Therapy in Sweden (ARTIS), the practice-based national Swedish Biologics Register. For each initiated treatment, information on the underlying rheumatologic condition, including the date of onset, date of treatment initiation (and discontinuation), type and dose of biologic agent, scores on the Disease Activity Score 28-joint assessment (DAS28) and the Health Assessment Questionnaire (HAQ), concomitant disease-modifying antirheumatic drugs (DMARDs), steroids, nonsteroidal antiinflammatory drugs, and analgesics, is recorded by the treating rheumatologist at the start of treatment as well as at prespecified followup visits. By linking the above-mentioned national cohort of 67,743 RA patients to the Swedish Biologics Register, we identified those who were starting anti-TNF therapy, including their dates of starting/stopping.
A total of 6,604 patients with RA started therapy with their first TNF antagonist between February 1998 and July 31, 2006 (Table 1). The first anti-TNF agent was infliximab in 51%, etanercept in 34%, and adalimumab in 14%. Twenty-five percent of patients had been treated with >1 anti-TNF agent. The distribution of patient characteristics (Table 1) among the etanercept, infliximab, and adalimumab groups was strikingly similar, apart from a higher proportion of concomitant use of methotrexate (MTX) among patients starting infliximab (Table 1).
|All first anti-TNFα (n = 6,604)||First etanercept (n = 2,287)||First adalimumab (n = 937)||First infliximab (n = 3,380)|
|Males||1,630 (25)||520 (23)||216 (23)||894 (26)|
|Females||4,974 (75)||1,767 (77)||721 (77)||2,486 (74)|
|Age at start of anti-TNFα, mean years||55||54||55||55|
|Year of first anti-TNFα start|
|1998–2001||2,491 (37)||939 (41)||0 (0)||1,552 (46)|
|2002–2003||1,824 (28)||473 (21)||254 (27)||1,097 (32)|
|2004–2006||2,289 (35)||875 (38)||683 (73)||731 (22)|
|RA duration at start of anti-TNFα, mean years||10.6||10.7||11.9||10.1|
|Joint counts at start of anti-TNFα (28 assessed)|
|Mean no. of swollen joints||9.5||9.8||9.1||9.4|
|Mean no. of tender joints||8.7||9.1||8.4||8.6|
|Mean DAS28 score at start of anti-TNFα||5.5||5.6||5.4||5.5|
|Mean HAQ score at start of anti-TNFα||1.4||1.4||1.3||1.3|
|Mean ESR at start of anti-TNFα, mm/hour||37||38||34||37|
|DMARDs at start of anti-TNFα|
|None||1,057 (16)||724 (32)||190 (20)||143 (4.2)|
|MTX||4,563 (69)||1,194 (52)||605 (65)||2,764 (82)|
|Other than MTX||697 (11)||285 (13)||113 (12)||299 (8.8)|
|No information||287 (4)||84 (3.7)||29 (3.1)||174 (5.1)|
|Yes||3,383 (51)||1,309 (57)||522 (56)||1,552 (45)|
|No||2,216 (33)||732 (32)||369 (40)||1,115 (33)|
|No information||1,005 (15)||246 (11)||46 (4.9)||713 (21)|
|Comorbid conditions at start of anti-TNFα|
|Any infection||746 (11)||279 (12)||115 (12)||352 (10)|
|Diabetes||268 (4.1)||105 (4.6)||27 (2.9)||136 (4.0)|
|Any cardiovascular disease||478 (7.2)||180 (7.8)||73 (7.8)||225 (6.7)|
|Any COPD||228 (3.5)||93 (4.1)||25 (2.7)||110 (3.3)|
|Joint replacement surgery||1,852 (28)||696 (30)||256 (27)||900 (27)|
|History of cancer||238 (3.6)||71 (3.1)||37 (3.9)||130 (3.8)|
Linkage of all unique individuals mentioned above to the Swedish Cancer Register (to which reporting of all incident malignancies has been mandatory, both for the treating physician and for the pathologist, since 1958, and resulting in a near-complete coverage of all diagnosed cancers in Sweden ) identified all registered cancers from 1958 through December 2006 for all unique individuals in the study. Validation with chart review of >50% of all incident cancers registered among the RA patients exposed to an anti-TNF agent in our study indicated >99% of the events were de facto occurrences of cancer of the registered type, and were diagnosed after the start of anti-TNF treatment.
Through their personal identification numbers, all unique 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). A total of 0.3% of all unique individuals were excluded because of data inconsistencies that precluded unambiguous followup throughout the period of our study, which started February 1, 1998 at the earliest and ended December 31, 2006 at the latest.
To assess the RR of cancer in patients with RA who were taking an anti-TNF agent as compared with biologics-naive RA patients, we used Cox regression to perform an internal cohort analysis within the national cohort of 67,743 RA patients. Anti-TNF treatment was treated as a time-dependent variable. In other words, patients in the RA cohort who proceeded to treatment with a TNF antagonist at some stage after the start of followup were censored from the anti-TNF–naive group at this time point and contributed to the anti-TNF–treated group thereafter. The Cox models used calendar time as the time scale and were stratified by sex, county of residence, and year of birth in 1-year strata, such that the RRs were assessed while controlling for attained age and calendar time.
Analyses were adjusted for combinations of the following: marital status and time-dependent comorbidity hospitalization covariates (hip, knee, shoulder, and ankle joint replacement, diabetes, cardiovascular disease, infection, chronic obstructive pulmonary disease, cumulative number of hospitalizations, cumulative number of RA hospitalizations, cumulative number of days spent in hospital, and duration of RA, as well as the RA cohort source[s] from which each patient was recruited) until 1 year before the occurrence of the cancer. Adjustment for these factors changed the RR estimates by ∼5%, and these adjusted RRs are presented below. Because of the small number of events in many strata, overparameterization was a concern. However, considerably less-stratified models resulted in very similar risk estimates.
Separate RRs were estimated according to sex, age, calendar year of anti-TNF start, time since anti-TNF treatment start, and duration of anti-TNF therapy. In all analyses, patients with a history of any cancer at the start of followup were excluded, and followup was censored at the diagnosis of the first primary cancer, irrespective of tumor type or site. Cancer risks for all sites were assessed twice, with and without including nonmelanoma skin cancers (which in the present study, reflects squamous cell skin cancer, since basal cell skin cancer was not reported to the Cancer Register during the study period).
To assess the impact of comparing new-starters of anti-TNF therapy with contemporary new starters of nonbiologic agents (rather than, as above, comparing them with all other RA patients in Sweden), we identified all patients in the Swedish Early RA Register who started a second nonbiologic DMARD, including their date of starting this therapy (n = 1,838), from 1998 through 2006 (mean DAS28 at the start of therapy 4.5). Similarly, we identified all patients in this register who were first starting MTX during the same time period (n = 5,989; mean DAS28 at the start of therapy 5.0). Through use of Cox regression, the cancer incidence in the ARTIS cohort of 6,604 anti-TNF starters was compared with each of these new-start DMARD cohorts, using the same adjustments and stratifications as in the original Cox regression analyses described above, plus the duration of RA. As a sensitivity analysis, we compared in a similar manner only those patients in the ARTIS anti-TNF cohort who were also members of the Early RA cohort with the new starters of MTX and a second DMARD, respectively, as well as with all other members of the Early RA register.
Using a similar Cox regression analysis, we also compared the occurrence of cancer overall and by time since the start of therapy in the anti-TNF–treated cohort with that in the general population, using a general population comparator of 408,048 randomly selected subjects who were drawn from the Swedish Population Register and matched to each unique RA patient in the national RA cohort by sex, age, county of residence, and marital status.
Counting from the start of the first anti-TNF agent until the patient's death, emigration, date of diagnosis of the first cancer, or end of the study period, 240 first primary cancers were diagnosed during 25,693 person-years of followup of the 6,366 patients starting anti-TNF therapy who had no history of cancer at the start of anti-TNF treatment. Table 2 displays the characteristics of this cancer occurrence.
|All first anti-TNFα (n = 6,366)||First etanercept (n = 2,216)||First adalimumab (n = 899)||First infliximab (n = 3,249)|
|Person-years of followup|
|<1 year since start of anti-TNFα||6,310||2,199||887||3,223|
|≥1–2 years since start of anti-TNFα||5,738||1,989||739||3,008|
|≥2–4 years since start of anti-TNFα||7,951||2,587||530||4,832|
|≥4–6 years since start of anti-TNFα||4,310||1,683||2.6||2,670|
|≥6 years since start of anti-TNFα||1,387||999||0||387|
|No. of first cancers||240||70||26||144|
|Crude incidence (range per 100,000)||934 (819–1,060)||743 (580–939)||1,204 (786–1,763)||1,020 (860–1,201)|
|Age/sex standardized incidence†||934||749||1,238||1,002|
|Stage at cancer diagnosis, no. (%) of cancers‡|
|Total (% of all cancers)||115 (48)||31 (56)||12 (54)||72 (50)|
|Stage I (% of all with information)||51 (44)||14 (45)||5 (42)||32 (44)|
|Stage II||36 (31)||10 (32)||3 (25)||23 (32)|
|Stage III||14 (13)||2 (6.5)||4 (33)||8 (22)|
|Stage IV||14 (12)||5 (16)||0 (0)||9 (13)|
|Site, no. (%) of cancers|
|Respiratory tract||36 (15)||9 (13)||2 (7.7)||25 (17)|
|Gastrointestinal tract||31 (13)||8 (11)||4 (15)||19 (13)|
|Reproductive tract||75 (31)||25 (35)||10 (39)||40 (28)|
|Urogenital tract||15 (6.3)||8 (11)||1 (3.9)||6 (4.2)|
|Skin||32 (13)||7 (10)||2 (7.7)||23 (16)|
|Hematopoietic system||28 (12)||8 (11)||4 (15)||16 (11)|
|Other sites||22 (9.2)||5 (7.1)||3 (12)||14 (9.7)|
Figure 1 displays the incidence rate of cancer according to the time since the start of anti-TNF therapy. Taking age and sex into account, we found no trend toward an increasing or decreasing incidence of cancer with an increasing duration of followup (P for linear trend = 0.36).
A total of 240 incident first primary cancers occurred among 6,366 patients with RA who had started anti-TNF therapy. Comparing these first primary cancers with the 4,244 first primary cancers that occurred during 330,498 person-years of followup among 61,160 anti-TNF–naive patients in the national RA comparator cohort of patients who did not have a history of cancer at start of the study, the relative risk was 1.00 (95% confidence interval [95% CI] 0.87–1.17) (Table 3). This relative risk remained largely similar across different time intervals since the first anti-TNF start (Table 3) and did not change with increasing cumulative time receiving active anti-TNF therapy (e.g., RR = 0.96 [95% CI 0.50–1.86] for those with ≥6 years of anti-TNF therapy) (Table 3). There was no obvious heterogeneity in the RRs across the categories of sex or attained age.
|Anti-TNFα–treated patients||Anti-TNFα–naive patients||RR (95% CI)|
|Overall number of cancers||240||4,244||1.00 (0.87–1.17)|
|Sex, no. of cancers|
|Age at start of anti-TNFα therapy|
|<50 years||21||142||1.21 (0.68–2.16)|
|≥50–74 years||179||2,216||0.98 (0.83–1.16)|
|≥75 years||40||1,886||1.03 (0.73–1.50)|
|Time since start of anti-TNFα therapy|
|<1 year||56||763||1.04 (0.77–1.39)|
|≥1–2 years||40||632||0.83 (0.59–1.16)|
|≥2–4 years||83||1,182||1.11 (0.88–1.41)|
|≥4–6 years||46||817||0.96 (0.70–1.32)|
|≥6 years||15||850||1.00 (0.58–1.72)|
|Cumulative duration of anti-TNFα therapy|
|<2 years||136||–||1.00 (0.83–1.22)|
|≥2–4 years||55||–||0.87 (0.65–1.15)|
|≥4–6 years||39||–||1.23 (0.88–1.73)|
|≥6 years||10||–||0.96 (0.50–1.86)|
|Year of first anti-TNFα start|
When assessed separately for each of the 7 groups of organ system–specific cancers listed in Table 2, the RR for first primary cancers associated with anti-TNF therapy ranged from 0.9 to 1.4, with none of the estimates attaining formal statistical significance (data not shown). When nonmelanoma (squamous cell) skin cancers were excluded, the overall RR was 0.96 (95% CI 0.82–1.11), and no trend for the duration of followup was revealed.
Comparison of the 240 first primary cancers among the 6,366 patients with RA who had started anti-TNF with the 260 cancers among the 5,989 RA patients newly starting MTX therapy yielded an RR of 0.99 (95% CI 0.79–1.24) (Table 4). When this comparison was restricted to those anti-TNF–treated patients who were also part of the Early RA Register (1,632 patients; 38 cancers) so that an internal analysis within the Early RA cohort was performed, no increased risk was observed (RR 0.96 [95% CI 0.64–1.42]). When the 240 first primary cancers among the 6,366 patients with RA who had started anti-TNF were compared with the 64 cancers among the 1,838 patients newly starting a second nonbiologic DMARD, the relative risk was 0.97 (95% CI 0.69–1.36) (Table 4).
|RR (95% CI) of first primary cancer in 6,366 RA patients starting anti-TNFα therapy versus|
|General population (n = 408,048)||RA patients starting MTX therapy (n = 5,989)||RA patients starting nonbiologic DMARD combination therapy (n = 1,838)|
|No. of person-years in comparator group||3,094,878||23,558||7,043|
|No. of cancer cases in comparator group||30,490||260||64|
|RR (95% CI)|
|Overall||1.14 (1.00–1.30)||0.99 (0.79–1.24)||0.97 (0.69–1.36)|
|Time since start of anti-TNFα|
|<1 year||1.17 (0.90–1.53)||1.08 (0.76–1.55)||0.99 (0.62–1.58)|
|≥1–2 years||0.90 (0.66–1.23)||0.82 (0.56–1.21)||0.78 (0.49–1.26)|
|≥2–4 years||1.25 (1.01–1.56)||1.11 (0.83–1.49)||1.13 (0.76–1.69)|
|≥4–6 years||1.19 (0.89–1.59)||0.87 (0.60–1.27)||0.86 (0.54–1.38)|
|≥6 years||1.20 (0.72–2.01)||1.04 (0.55–1.98)||1.14 (0.56–2.32)|
When each of the 3 groups of first starters of the 3 anti-TNF agents was separately compared with the biologics-naive national RA cohort, none of the 3 anti-TNF agents was associated with an increased overall risk of cancer (Table 5). Exclusion of nonmelanoma skin cancer did not appreciably alter these RRs (data not shown). When the overall relative risks were further explored, dissimilar risks were observed during the first year of followup: first starters of etanercept had a lower risk of cancer during the first year of followup, as compared with the comparator cohort of unselected, biologics-naive contemporary patients with RA, whereas patients first starting adalimumab were at an increased risk during the first year after starting.
|RR (95% CI) and number of events, by first anti-TNFα agent received||All first anti-TNFα therapy as a single class (n = 6,364)|
|First etanercept (n = 2,216)||First infliximab (n = 3,249)||First adalimumab (n = 899)||P†|
|Overall||0.78 (0.61–1.00); 70||1.09 (0.91–1.30); 144||1.32 (0.87–1.98); 26||0.034||1.00 (0.86–1.15); 240|
|Time since start of anti-TNFα|
|<1 year||0.43 (0.22–0.84); 10||1.23 (0.85–1.77); 31||1.91 (1.11–3.31); 15||0.0027||1.03 (0.78–1.36); 56|
|≥1–2 years||0.80 (0.46–1.40); 13||0.83 (0.53–1.28); 21||0.84 (0.37–1.92); 6||0.99||0.82 (0.59–1.13); 40|
|≥2 years||0.92 (0.68–1.24); 47||1.13 (0.91–1.41); 92||1.08 (0.43–2.67); 5||0.53||1.05 (0.88–1.25); 144|
Beyond the first year of followup, no difference between the 3 drugs was observed, with all RR estimates close to 1 (Table 5). Censoring followup at the time of switching from the first to the second anti-TNF agent did not result in any marked alteration of these RRs (data not shown).
In this national and population-based assessment of cancer occurrence among Swedish RA patients treated with TNF antagonists in clinical practice, we made the following 3 main observations. First, overall and during the first years following the start of anti-TNF therapy, patients treated with anti-TNF had a cancer risk that was similar to that of anti-TNF–naive RA patients, that of patients starting MTX therapy, as well as that of patients starting DMARD combination therapy. Second, neither the incidence nor the relative risk of cancer increased with time since first starting anti-TNF therapy, nor did they increase with the cumulative duration of active anti-TNF therapy. Third, although none of the 3 anti-TNF agents was itself associated with an increased overall risk of cancer, each of them displayed somewhat different cancer risks during the first year of followup, but not thereafter.
Randomized controlled trials (RCTs) provide balanced groups for analysis, thereby satisfying one of the prerequisites for a valid comparison of cancer risk; however, certain other features of RCTs limit the amount of information that can be deduced. With an average duration of only a few months, RCTs provide little insight into the time periods during which an effect of anti-TNF therapy on cancer risk would be more likely to manifest itself. Moreover, although “cancer” has a superficially straightforward definition and is a serious adverse event, retrospective identification and adjudication of cases has proved a methodologic challenge (7). Finally, although meta-analytical approaches increase statistical precision, the overall number of cancers in these trials is still modest, particularly in the comparator arms, which are often only half the size of the anti-TNF arm (5, 6). Nevertheless, from some of these studies, signals have emerged to indicate that there may be an increased risk of cancer with all 3 anti-TNF agents within the first few months of treatment (5, 6).
Observational studies allow for both a greater number of (less selected) study subjects and a longer followup time but are critically dependent upon the choice of the comparator and the identification of “cancers.” So far, observational studies have not indicated any increased overall risk of cancer, although again, the time frames covered have been rather short (mean followup in all studies ≤3 years) (10–12, 21), and the number of cancer events in the anti-TNF groups has been moderate (n = 67 , n = 231 cancers other than nonmelanoma skin cancer , n = 57 , and n = 16 ).
So far, only 1 observational study, the one reported by Wolfe and Michaud (12), has presented RRs for cancer per individual anti-TNF agent. In that study, ever exposure to anti-TNF agents was not associated with an increased cancer risk (RR 1.0, based on 231 cancers other than nonmelanoma skin cancer among the biologics-exposed patients), nor was ever exposure to any of the 3 agents separately (range of RRs 0.7–1.0). Although no trend toward a higher RR with increasing duration of ever exposure to anti-TNF therapy was observed in that study, patients were not followed up from the start of anti-TNF therapy, and this precluded any assessment of risk per time since treatment start, as well as any comparison of risks by each of the 3 anti-TNF drugs as a function of time of followup.
In the study by Setoguchi et al (11), an overall RR of 0.98 (95% CI 0.73–1.31) for the development of cancer in patients taking TNF antagonists was reported. Two Swedish studies, both based on data that were included in the current analysis, have previously assessed cancer risk in relation to anti-TNF therapy. In the study by Geborek et al (21), a geographic subset of the Swedish Biologics Register was compared with the general population rates for cancer from 1999 through 2002. The results suggested an overall RR of 1.1 (95% CI 0.6–1.8) (21). In the other Swedish study, a preliminary assessment of cancer risk using national data from the Swedish Biologics Register 1999–2003 and cancer data through 2003, an RR of 0.9 (95% CI 0.7–1.2) was reported (10).
Therefore, our observation that the overall risk of cancer was unaltered is consistent with the findings of previous observational studies. With respect to our second observation that there were no trends toward increasing (or decreasing) risk over time, there are still few comparison data, and a number of caveats (see below) should be borne in mind. Our third observation is the finding of a difference in cancer risk between the 3 anti-TNF agents during the first year but not during subsequent years. To our knowledge, our study is the first to perform this type of comparison, and it raises 2 distinct questions: Is it plausible that any anti-TNF agent would increase the risk of cancer in only 12 months, a relatively brief time span for carcinogenesis? And, is it plausible that a true biologic difference exists between adalimumab, etanercept, and infliximab with regard to the occurrence or diagnosis of cancer?
The effect of an anti-TNF agent on cancers seen during such an early stage of therapy might be due to existing preclinical tumors becoming clinically manifest, either by allowing more actual tumor growth or by influencing the development of clinical symptoms as a result of the presence of the cancer, rather than being due to the occurrence of a new cancer. Because of subtle differences in the exact biologic properties of adalimumab, etanercept, and infliximab, we cannot completely exclude the possibility that there is a true biologic difference between the 3 anti-TNF agents relevant to the occurrence of cancer. However, it must also be emphasized here that the circumstances under which the 3 anti-TNF agents have been used in Sweden were somewhat dissimilar and that subtle influences of patient selection for each of the 3 anti-TNF agents may further modulate any risk of cancer or its diagnosis. Consequently, the differences observed in our study should be interpreted with great caution, and additional studies are needed in order to determine whether the observed differences are either artifactual or chance events or whether they mirror true biologic differences. Importantly, after the first year of followup, no appreciable between-drug difference was observed, suggesting that any true difference might be related more to cancer detection than to cancer causation.
The strengths of our study include the population-based design, with inclusion independent of insurance status, income, or referral patterns, and the use of population-based registers to identify RA patients, treatments, and cancers. The identification of incident cancer at any time from the start of treatments allowed the assessment of risk as a function of time since the start of therapy and minimized the risk that any early effects on cancer risk (leading to treatment discontinuation) would be overlooked. Similarly, the use of longitudinal data on anti-TNF treatment status permitted the assessment of risk as a function of the duration of active anti-TNF therapy, rather than merely the time since the agent was first started. The use of multiple comparators provided the possibility to identify any signals driven spuriously by aberrant rates in 1 of the comparators rather than in the anti-TNF group itself. Importantly, comparison with patients starting nonbiologic therapy for RA, rather than with unselected patients with RA, allowed for the assessment of risks that may be influenced by the therapeutic context (i.e., pretreatment evaluation, general vigilance following start of a new long-term therapy, etc.) (13).
The followup of our study cohorts was virtually complete, with <1% of the population being lost to followup. Cancer was identified through linkage of the study population to the Swedish Cancer Register, to which reporting is mandatory and semiautomated for all clinicians, resulting in an exceedingly high completeness (>95% [14, 20]). The use of this external and independent source based on mandatory reporting for the assessment of cancer decreased the risk of uneven reporting rates for the anti-TNF group versus its comparators, as well as the risk of time trends driven by a decreasing propensity for reporting, such as may occur within the context of, for example, a long-term safety study that relied on rheumatologists' reporting of outcomes. Moreover, validation of the cancer diagnosis in the anti-TNF group confirmed its high diagnostic accuracy.
Certain limitations of our study should be mentioned. Although this study currently represents the largest assessment of cancer risk following initiation of anti-TNF therapy, precision in several of the analyses is still limited, particularly in the assessment of drug-specific RRs by time since the start of treatment. For the same reason, the aim of this study was restricted to the assessment of overall, rather than site-specific, cancer risk. Although the unaltered overall RR is reassuring, altered RRs for cancers at less common sites may have been overlooked. Despite our comparatively long followup (median 3.6 years; maximum 8 years), the followup may not be long enough to detect insidious effects of anti-TNF therapy occurring early in carcinogenesis.
Despite the use of multiple comparators and despite adjustments for comorbid conditions and for patient demographics, uncontrolled channeling bias is a constant threat to the validity of observational studies. Channeling of patients to anti-TNF treatment may have resulted in a group of patients who are at increased risk of developing cancer (e.g., high levels of inflammatory activity, treatment with other immunosuppressive agents, etc.) as compared with unselected patients. In contrast, pretreatment evaluations that include chest radiographs and certain laboratory tests may incidentally detect subclinical cancers before anti-TNF treatment is begun (and which is then never begun); thus, cancers that would otherwise have been diagnosed shortly after the start of anti-TNF treatment may thereby have been depleted from the cohort. Similarly, although we had detailed information on treatments and cancers, we had limited information on other risk factors for cancer, and such risk factors may differ between patients starting therapy with a biologic agent and those starting DMARD or combination therapy.
Clinical selection of patients to receive 1 of the 3 anti-TNF agents might also have introduced differences in the true risk factor profile, despite seemingly similar RA and comorbidity characteristics. The treatment setting may alter the probability of detecting a subclinical cancer. For example, patients starting infliximab therapy are likely to have 7–8 more annual health contacts than are patients starting a self-administered therapy, which would open it up for detection bias. One would assume that earlier detection of cancers (“lead-time bias”) would be accompanied by a more favorable distribution of cancer stages at diagnosis. In this study, we compared the stage at diagnosis between the overall group of patients taking anti-TNF agents (versus the biologics-naive RA cohort) as well as for each of the 3 anti-TNF agents, but no major differences were identified. However, information on tumor stage was only available for a subset of all cancers, and any comprehensive assessment of cancer stage distribution needs also to take into account other covariates, such as age, sex, and the distribution of cancer sites (Table 1).
With regard to generalizability of our findings, the population-based setting would suggest that there is modest room for selection bias. However, the generalizability to other treatment settings is not guaranteed; other risk factors for cancer (e.g., exposure to ultraviolet light, lifestyle factors, and body mass index) may vary between populations and might modify any effect of anti-TNF therapy and cancer. In our national RA cohort, the prevalence of anti-TNF therapy was ∼15%. It cannot be excluded that in the treatment setting with considerably lower disease severity, cancer risks with anti-TNF therapy would be different.
We conclude that the overall occurrence of cancer during the first years following anti-TNF therapy in RA is not higher than that in biologics-naive patients with RA, nor does it increase with time. During the first year of followup, but not thereafter, the 3 anti-TNF agents displayed somewhat different cancer risks, the reason for which is unclear. At present, and given the remaining uncertainties, continued vigilance remains prudent.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Askling had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Askling, van Vollenhoven, Fored, Cöster, Geborek, Lindblad, Saxne, Klareskog.
Acquisition of data. Askling, van Vollenhoven, Raaschou, Dackhammar, Feltelius, Cöster, Geborek, Jacobsson, Lindblad, Rantapää-Dahlqvist.
Analysis and interpretation of data. Askling, van Vollenhoven, Granath, Baecklund, Saxne, Klareskog.
The authors were in charge of, and solely responsible for, all data collection, analysis, and interpretation, the writing of the manuscript, and the decision to submit the manuscript for publication, without any constraints exerted by the agencies or companies that provided financial support for the ARTIS Register. Publication of this article was not contingent upon approval by any agency or sponsor.
We gratefully acknowledge the assistance of Maud Rytting (Medical Products Agency) and the Swedish RA Register/ARTIS representatives at the following centers for allowing us to use their data: Yngve Adolfsson (Sunderby Hospital, Luleå), Ewa Berglin (Norrland University Hospital, Umeå), Torgny Smedby (Östersund County Hospital), Rüdi Götze (Sundsvall County Hospital), Anna-Carin Holmqvist (Hudiksvall Hospital), Sven Tegmark (Gävle County Hospital), Jörgen Lysholm (Falu Lasarett, Falun), Solveig Gustafsson (Karlstad Central Hospital), Eva Baecklund (Akademiska Hospital, Uppsala), Rolf Oding (Västerås Hospital), Per Salomonsson (University Hospital, Örebro), Birgitta Nordmark (Karolinska University Hospital, Solna), Ingiäld Hafström (Karolinska University Hospital, Huddinge), Göran Lindahl (Danderyd Hospital, Stockholm), Gun Sandahl (Queen Sophia Hospital, Stockholm), Martin Mousa (Visby Lasarett, Visby), Anders Lindblad (Visby Privat, Visby), Åke Thörner (Mälarsjukhuset, Eskilstuna), Lars Cöster (University Hospital, Linköping), Sören Transö (County Hospital Ryhov, Jönköping), Olle Svernell (Västervik Hospital), Claudia Jacobs (Oskarshamn Hospital), Bengt Lindell (Kalmar County Hospital), Maria Söderlin (Växjö Central Hospital), Olof Börjesson (Växjö Privat), Göran Kvist (Centrallasarettet Borås), Karin Svensson (Kärnsjukhuset, Skövde), Tomas Torstenson (Uddevalla Hospital), Ingeli Andreasson (Göteborg Privat), Lennart Bertilsson (Sahlgrenska University Hospital, Gothenburg), Tore Saxne (University Hospital Lund), Miriam Karlsson (Lasarettet Trelleborg), Annika Teleman (Spenshult, Oskarström), Catharina Keller (Helsingborg Lasarett), Astrid Schröder (Ängelholm Hospital), Jan Theander (Kristianstad Central Hospital), and Christina Book (MAS University Hospital, Malmö).