Objective To determine whether tumour necrosis factor inhibitor (TNFi) use is associated with an increased rate of incident malignancy compared with no TNFi use in the treatment of juvenile idiopathic arthritis (JIA), paediatric inflammatory bowel disease (pIBD) and paediatric plaque psoriasis (pPsO).
Methods We performed a retrospective cohort study of administrative claims data from the USA from 2000 to 2014. Exposure to TNFi was considered permanent from the first observed exposure onward. The malignancy outcome was defined by diagnosis codes with evidence of cancer treatment. We calculated standardised incidence ratios (SIRs) comparing the observed number of malignancies to the expected numbers according to cancer surveillance data. We used multivariable Cox proportional hazards models to estimate adjusted HRs (aHRs) for incident malignancy.
Results We identified 15 598 children with TNFi use and 73 839 children with no TNFi use (30 703 and 121 801 person-years of follow-up, respectively). We identified 15 malignancies among children with TNFi use (SIR 2.9 (1.6 to 4.9)) and 42 malignancies among children without TNFi use (SIR 2.1 (1.5 to 2.9)). The aHR was 1.58 (0.88 to 2.85) for TNFi use versus no TNFi use. In pIBD, TNFi use with thiopurine use was associated with a higher SIR (6.0 (1.2 to 17.5)) compared with TNFi use without thiopurine use (2.5 (0.7 to 6.4)).
Conclusion Children diagnosed with JIA, pIBD and pPsO had an increased rate of malignancy compared with the general population, but treatment with TNFi did not appear to significantly further increase the risk compared with no TNFi use. More data are needed about the long-term risks of TNFi use.
- Juvenile Idiopathic Arthritis
Statistics from Altmetric.com
Tumour necrosis factor inhibitors (TNFi) are highly effective for the treatment of juvenile idiopathic arthritis (JIA)1 2; paediatric inflammatory bowel disease (pIBD), including Crohn’s disease3 4 and ulcerative colitis5 ; and paediatric plaque psoriasis (pPsO).6 Indeed, the advent of TNFi began the era of treatment with biologic agents, which has greatly improved expected outcomes in these chronic conditions.
Nevertheless, there is significant worry about the potentially increased rate of malignancy associated with paediatric use of TNFi compared with the rate in the general population, as first reported by the United States Food and Drug Administration (FDA) in 2009.7 This initial report had many limitations, including failure to account for a possible increased risk of malignancy associated with the underlying conditions being treated with TNFi (ie, increased background risk of malignancy) or for a possible increased risk associated with other immunosuppressive medications, such as the thiopurines.8
To date, the very low incidence of paediatric malignancy and the relatively small number of children exposed to TNFi have limited attempts to definitively prove or disprove an independent association between paediatric TNFi use and subsequent malignancy. In this study, we used administrative claims data from the USA to determine the comparative rates of malignancy among children with JIA, pIBD and pPsO who were and were not treated with TNFi.
After obtaining institutional review board approval, we performed this study using national US Medicaid Analytic eXtract files (MAX) (from government-provided health insurance for low-income families) from 2000 through 2010 and national US Truven MarketScan files (from commercial health insurance via employers) from 2010 through 30 September 2014. Race and ethnicity data were not available in MarketScan. We used all of the data available to us at the time of the study.
We identified cohorts of children with the most common disease indications for TNFi. Criteria for study inclusion were (a) the first physician diagnosis code for pPsO, JIA or pIBD prior to age 18 years; and (b) at least two physician diagnosis codes for pPsO, JIA or pIBD that were >7 and <183 days apart OR any one physician diagnosis code for pPsO, JIA or pIBD that was followed by a prescription claim for immunosuppressant medications typically used to treat pPsO, JIA or pIBD (eg, methotrexate, azathioprine, TNFi) within 183 days. Patients were excluded for diagnosis codes of pPsO, JIA or IBD prior to 6 months of age and for any diagnosis codes for organ transplantation, HIV infection (HIV) or systemic lupus erythematosus (SLE) and related rheumatological conditions prior to the start of follow-up. Follow-up for each patient began 183 days after the first diagnosis code of the codes that satisfied criterion (b) above in order to exclude prevalent or misdiagnosed malignancy. Patients were excluded for any diagnosis codes for malignancy or any claims for cancer chemotherapy prior to the start of follow-up. Patients were assigned to disease cohorts using a hierarchy (pPsO<JIA<pIBD) and could change during the course of the study (eg, a patient’s diagnosis could change from JIA to pIBD, but not the reverse).
We used the same approach to identify a cohort of children diagnosed with attention-deficit hyperactivity disorder (ADHD) to evaluate the performance of our malignancy outcome algorithm, described below. In addition to the exclusions above, children were excluded from the ADHD cohort for any diagnoses of pPsO, JIA or IBD or any exposure to any of the medications of interest at any time.
We identified exposure to any of the five commercially available TNFi (adalimumab, certolizumab, etanercept, golimumab, infliximab) using pharmacy and infusion claims, and all TNFi were considered together as a single exposure group. For patients with pIBD, exposure to thiopurines (azathioprine and mercaptopurine) was also assessed. For both TNFi and thiopurines, patients were considered permanently exposed from the first observed exposure onward, including exposures prior to the start of study follow-up. Exposure categories were binary, and patients could contribute follow-up time to more than one medication exposure group sequentially throughout the study according to their treatment course (eg, a patient could begin follow-up in the no TNFi use group and switch to the TNFi group on first observed exposure). We classified new TNFi users as defined by >6 months of observation without any TNFi exposure prior to the first observed TNFi exposure.
We used an adapted version of a malignancy-finding algorithm that was previously validated in adults.9 10 Incident malignancy was defined by at least one physician diagnosis code for malignancy accompanied by claims evidence of treatment for malignancy (ie, chemotherapy, radiation therapy or surgical excision). Diagnoses of non-melanoma skin cancer (NMSC) were ignored because they could not be sufficiently substantiated.
Follow-up for all patients continued until one of the following events occurred: end of study period, malignancy outcome, loss of observability in the data (ie, loss of healthcare coverage benefits) or any diagnosis codes for organ transplantation, HIV or SLE.
We determined the malignancy rates for the disease cohorts stratified by exposure to TNFi. We then combined results across diseases. We used Surveillance, Epidemiology, and End Results (SEER) data to calculate the expected number of malignancies according to the age, sex and race distributions of follow-up time in each cohort. We calculated standardised incidence ratios (SIRs) with 95% CIs comparing the observed number of malignancies to the expected numbers according to SEER. We separately evaluated lymphomas because of particular concern about this malignancy type.7
To assess the performance of the malignancy outcome algorithm, we determined the rate of malignancy in the ADHD cohort and computed age, sex and race-adjusted SIR. The ADHD cohort is expected to have a rate of malignancy comparable to the general population.
We used Cox proportional hazards models to determine the HR for incident malignancy for TNFi use compared with no TNFi use. We assessed for confounding using bivariate models including age, sex, race and data source and included variables that altered the HR for TNFi use by approximately 10% or more. Because individual patients could contribute sequentially to both cohorts, a sandwich variance estimator was applied to account for additional correlations in the data.11 Analyses were performed using SAS V.9.4 (SAS Institute).
We identified 28 005 patients with JIA, 24 035 patients with pIBD and 31 438 patients with PsO, including 7419/6808/1371 TNFi users, respectively (table 1). The proportion of TNFi users was much greater among JIA (26%) and pIBD (28%) compared with PsO (4%). The proportion of TNFi users was higher among patients in MarketScan (27%) compared with MAX (13%). The median duration of follow-up after first TNFi exposure was 1.4 years, and 25% of TNFi users had at least 2.8 years of follow-up after TNFi exposure. The approximate distribution of TNFi use overall was etanercept 37%, infliximab 34%, adalimumab 29%, certolizumab 2% and golimumab 1%.
Table 2 shows the incident malignancy rates and corresponding SIR compared with age, sex and race adjusted SEER estimates. The incidence of malignancies in the ADHD cohort was highly consistent with that predicted by SEER, with a resultant SIR of 0.97 (0.91–1.05).
In total, there were 15 incident malignancies identified during 30 703 person-years of follow-up after TNFi use and 42 malignancies in 121 801 person-years of follow-up among those without TNFi use. The 15 malignancies following TNFi use included six lymphoma, three brain, two leukaemia, two malignant melanoma, one bone and one liver. The SIRs associated with no TNFi use for JIA, pIBD and pPsO were nearly identical (2.1 (1.1–3.5), 2.1 (1.1–3.6) and 2.1 (1.1–3.5), respectively). The SIR associated with TNFi use in JIA and pIBD was very similar (3.1 (1.3–6.1) and 3.3 (1.3–6.9), respectively), and there were no malignancies identified following TNFi use among those with pPsO (SIR 0 (0–9.2)). When the results for all three indications for TNFi were combined, the SIR associated with TNFi use was 2.9 (1.6–4.9) and the SIR associated with no TNFi use was 2.1 (1.5–2.9).
Table 3 shows the incident lymphoma rates and corresponding SIR. The incidence of lymphomas in the ADHD cohort was highly consistent with that predicted by SEER (SIR 1.05 (0.89–1.24)). In total, there were six lymphomas identified following TNFi use (three each for JIA and pIBD) and nine lymphomas among those without TNFi use (three for pIBD and six for pPsO). When all three indications for TNFi were combined, the SIR associated with TNFi use was 6.0 (2.4–14.5) and the SIR associated with no TNFi use was 2.7 (1.2–5.2).
Table 4 shows the results from Cox proportional HR models. The unadjusted HR for TNFi use versus no TNFi use was 1.45 (0.80–2.62). In bivariate models, this estimate was not confounded by age, sex or race. Adjusting for data source, the HR was 1.58 (0.88–2.85) for TNFi use versus no TNFi use. The HR for lymphoma was 2.64 (0.93–7.51) for TNFi use versus no TNFi use.
Of the 15 598 TNFi users, 9744 (62%) met the definition of new users and had 20 149 person-years (median 1.6 years) of follow-up after TNFi use. Among the TNFi new users, we observed nine incident malignancies. The median duration of time from new TNFi use to malignancy diagnosis was 343 days (range 38 to 1159), and three malignancies were diagnosed within 6 months of starting TNFi.
In the pIBD cohorts, children with thiopurine use and no TNFi use had two malignancies in 7770 person-years with an SIR of 1.5 (0.2–5.6). Children with TNFi use and no thiopurine use had four malignancies in 8913 person-years with an SIR of 2.5 (0.7–6.4). Children with TNFi use and thiopurine use had three malignancies in 2977 person-years with an SIR of 6.0 (1.2–17.5).
Despite nearly 20 years of use, there remains worry about the safety of TNFi, especially when used to treat children. This study using national administrative claims data from two sources in the USA assessed children with the three most common indications for TNFi and found no significant increased risk of overall malignancy associated with TNFi use compared with no TNFi use, although a 2-fold to 3-fold increase could not be excluded given the small number of observed malignancies. On the other hand, this study showed an approximate doubling of the incidence of malignancy associated with JIA, pIBD and pPsO in the absence of TNFi use compared with the general population. Taken together, these data suggest that, despite earlier reports to the contrary, TNFi use in childhood is not likely to substantially increase the overall risk of malignancy.
Our study confirmed previously published large studies showing an increased risk of malignancy associated with JIA in the absence of TNFi use. Large observational studies including those using linked national databases from Sweden12 and Taiwan13 and US claims data from MAX14 and commercial insurers15 have consistently found a 2-fold to 4-fold increase in malignancies in children with JIA in the absence of TNFi use compared with the general population.
There are fewer studies about the background risk of malignancy in pIBD and pPsO. Our study found an increase in incident malignancy associated with pIBD without TNFi use that was comparable with results from recently published large prospective16 and retrospective17 cohort studies of pIBD. Numerous previously published studies of adult IBD have shown an increased risk of several malignancies, particularly gastrointestinal cancers.18 Our study found a similarly increased incidence of malignancy associated with pPsO without TNFi use, and several studies of adult PsO have shown an increased risk of malignancy.19 20
Our study adds to the growing evidence that TNFi use is likely not strongly associated with subsequent malignancy. In JIA, our preliminary prior study did not identify any incident malignancies in 2922 person-years of follow-up after TNFi use.14 More recently, a report from the JIA German biologics registry did not observe an increase in the rate of malignancy following etanercept use compared with no TNFi use.21 The previously mentioned large prospective cohort of patients with pIBD evaluated malignancy following any biologic use (>95% of which was TNFi use).16 Biologic use without ever use of thiopurines was not associated with malignancy (SIR 1.11 (0.03–6.16)), while use of biologics with use of thiopurines was associated with malignancy (SIR 3.06 (1.32–6.04)). To our knowledge, there are no previously published large studies of the risk of malignancy associated with TNFi use in pPsO. In adults, large meta-analyses of clinical trials and large long-term observational studies have shown no increase in the risk of malignancy (aside from NMSC) following TNFi use in rheumatoid arthritis,22 23 IBD24 and PsO.25 26
One of the primary concerns in the initial FDA report was an increased incidence of lymphoma, up to 18 times the expected rate for children who received infliximab.7 Our study was limited by the few number of lymphomas identified. Nevertheless, the HR for lymphoma with TNFi use was much lower than the FDA’s result while still suggesting a possible true association (2.64 (0.93–7.51)). We could not adjust for high disease activity and severity, and these factors are believed to increase risk of lymphoma in adults with RA up to 70-fold compared with those with low disease activity.27
There are significant concerns about thiopurine use and malignancy, especially if TNFi are also used. Thiopurines alone have been shown to be associated with malignancy, particularly lymphoma, in adults with IBD.28 29 TNFi and thiopurines together have been shown to be higher risk than TNFi alone in adults with IBD.30 As noted above, TNFi and thiopurine use in pIBD was reported to have an approximately 2.8-fold increase in the SIR compared with TNFi use alone.16 Our study showed a similar approximately 2.4-fold increase in the SIR for TNFi and thiopurine use compared with TNFi use alone. The small number of malignancies and the typical prescribing patterns for TNFi prevented extensive evaluation of co-medications (eg, there were few children with JIA who received TNFi without having ever received methotrexate).
The beginning and end of the potential risk window for malignancy following TNFi initiation and discontinuation are unknown. We assumed an immediate risk of infinite duration on TNFi exposure to maximise the number of malignancies attributed to TNFi. When we restricted our analyses to new users of TNFi, we observed three of nine malignancies occurred within 6 months of TNFi initiation, a time period that may be too short to be appropriately attributed to TNFi. If these three cases are excluded from the overall TNFi use cohort, then the resultant SIR is reduced to 2.4 (1.2–4.1). The median time from initiation of TNFi to development of malignancy among the cases initially reported by the FDA was approximately 2.5 years.7 Greater than 25% of the TNFi users in this study had more than 2.5 years of observation after TNFi use, but overall long-term follow-up was limited.
A major challenge to the interpretation of studies of the association of TNFi and malignancy is confounding by indication. In studies of adults with RA27 and IBD,31 high disease activity and severity are directly associated with an increased rate of malignancy irrespective of treatment. Because high disease activity and severity are the very indications for paediatric use of TNFi, all observational studies are susceptible to bias towards an increased risk of malignancy associated with TNFi use. Methotrexate or thiopurines are often used prior to TNFi and may impart additional risks of malignancy. If one completely ignores the contribution of disease activity and prior treatments and attributes all of the observed increased risk of malignancy causally to TNFi use, the absolute increased rate of malignancy is still very small. We observed an absolute crude increased rate of malignancy of 14.4 per 100 000 person-years with a corresponding number needed to harm of 6944.
Our study had several important limitations. We did not have access to healthcare records to confirm the indications for TNFi or the diagnosis of malignancy. Overdiagnosis of the conditions commonly treated with TNFi among children without TNFi use would likely bias the study results against TNFi, given a background increased risk of malignancy associated with these conditions compared with the general population. Our malignancy outcome identification algorithm produced results for the ADHD cohort that were highly correlated with expected rates for overall malignancy and lymphoma. Also reassuring is the similarity between our results and the results of a prospective pIBD cohort in which pathological confirmation of malignancies was obtained.16 It is possible that some prevalent malignancies were misclassified as incident, but the use of a 6-month period to evaluate for prevalent malignancy prior to follow-up and the accuracy of our estimates for ADHD minimised this concern. Because NMSC is not reported to SEER, we could not evaluate the accuracy of that outcome and chose not to assess its incidence in our study. Owing to the few malignancy outcomes, we were unable to assess malignancy rates associated with individual TNFi. Because the rates of malignancy among children without TNFi use were so similar among the three diseases of interest, we chose to combine all patients into a single analysis to improve the precision of the estimates, but it is possible that there is effect modification by disease. We observed numerically different HRs for malignancy in our two data sources, although this may possibly be attributable to few observed outcomes, as the estimates had overlapping 95% CI. Lastly, the amount of observable follow-up after TNFi exposure was limited, largely because of frequent changes in healthcare coverage in the USA. A possible association between long-term TNFi use and malignancy could not be adequately evaluated. Many of these challenges may be partially addressed with long-term data from large prospective observational registries.16 21 32 33
In conclusion, our study demonstrates that being diagnosed with JIA, pIBD or pPsO increases the risk of incident malignancy and that use of TNFi does not appear to significantly further increase this risk in the first few years after use, with the possible exception of lymphoma.
Handling editor Josef S Smolen
Contributors TB conceptualised and designed the study, acquired the data, interpreted the results, drafted the initial manuscript, and reviewed and revised the manuscript. FX analysed the data and critically reviewed the manuscript. LC acquired the data, analysed the data and critically reviewed the manuscript. DBH, RM and MMM interpreted the results and critically reviewed the manuscript. JDL conceptualised and designed the study, interpreted the results and critically reviewed the manuscript. KGS acquired the data, interpreted the results and critically reviewed the manuscript. JRC conceptualised and designed the study, acquired the data, interpreted the results and critically reviewed the manuscript. All authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
Funding This study was funded by the United States Agency for Healthcare Research and Quality (AHRQ) as part of a grant (no U19HS021110) administered through the AHRQ Center for Education and Research on Therapeutics Program.
Competing interests TB has received consulting fees from UCB, Genentech/Roche, Novartis, Bristol-Myers Squibb and Sobi. JDL has received consulting fees from Takeda, Pfizer, Lilly, Gilead, Johnson & Johnson, Samsung Bioepis, AbbVie, UCB and Merck; and research funding from Takeda. RM has received consulting fees from Takeda. KGS has received consulting fees from Abbot, Amgen, Ardea/AstraZeneca, BMS, Crealta, Lilly, Merck, Pfizer and Genentech/Roche; and research funding from Amgen, Ardea/AstraZeneca, Crealta, Lilly, Merck and Takeda. JRC has received consulting fees from Genentech/Roche, UCB, Janssen, CORRONA, Amgen, Pfizer, BMS, Crescendo and AbbVie; and research funding from Genentech/Roche, UCB, Janssen, CORRONA, Amgen, Pfizer, BMS, Crescendo and AbbVie.
Ethics approval We obtained ethics approval from the Institutional Review Board of the University of Alabama at Birmingham, protocol number X140507011.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The MAX files and MarketScan files cannot be shared according to the terms of the data use agreements.
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.