Tumor Necrosis Factor Inhibitors and the Risk of Cancer among Older Americans with Rheumatoid Arthritis

Abstract

Background:

Tumor necrosis factor inhibitors (TNFis) effectively treat rheumatoid arthritis (RA) but may increase patient risk of some malignancies, particularly lymphomas or skin cancers.

Methods:

We used Surveillance, Epidemiology, and End Results (SEER)-Medicare data to conduct a case-control study in RA patients (2007–2015). Cases were individuals with a first cancer diagnosed in SEER registries (ages 66–99, 22 cancer sites, N=10,263). Skin cancer cases (non-melanoma skin cancer overall [NMSC, N=501]), basal cell carcinoma [BCC, N=161], squamous cell carcinoma [SCC, N=150]) and cancer-free controls (N=30,475) were selected from Medicare beneficiaries residing in SEER areas. Cases and controls had prior Medicare claims-based evidence for RA, and TNFi exposure was ascertained from part B and part D claims. Logistic regression was used to estimate adjusted odds ratios (aORs).

Results:

TNFi exposure was present in 16.2% of controls and 12.8–33.7% of cancer cases, varying by site. TNFi use was associated with increased risk of NMSC overall (aOR 1.32, 95%CI 1.06–1.63), non-Hodgkin lymphoma (NHL) overall (1.28, 1.06–1.56) and, specifically, follicular lymphoma (2.63, 1.63–4.24). TNFi exposure was not associated with other SEER cancer sites, BCC or SCC specifically, or other common NHL subtypes.

Conclusions:

Among older adults with RA, TNFi exposure was associated with elevated risk of NMSC and NHL, driven specifically by follicular lymphoma. Exposure was not associated with increased risk for other cancer sites.

Impact:

Our results support a role for tumor necrosis factor in lymphomagenesis. Given the association with NMSC, patients initiating TNFi therapy may benefit from skin cancer screening and sun protection measures.

Background

Rheumatoid arthritis (RA) is an autoimmune disease that causes inflammation and pain in the joints. Globally and in the United States (US), approximately 0.5% of the general population have RA.^1,2 Tumor necrosis factor (TNF), an important cytokine in the inflammatory cascade, is highly expressed in RA-affected joint tissue, and TNF is thus an important therapeutic target for RA.^3,4 TNF inhibitors (TNFis) are effective in controlling active RA and preventing progressive joint damage.⁴ TNFis are also used to treat other autoimmune or inflammatory conditions including inflammatory bowel disease (IBD, i.e., Crohn’s disease, ulcerative colitis).

Despite their utility in treating RA and other conditions, TNFis are associated with increased risk of serious infections, notably the reactivation of latent tuberculosis.^5–7 Because of the apparent immunosuppressant effects of TNFis, there have also been concerns that they could increase malignancy risk, especially for cancers that have infectious or immune-related etiology. This issue is especially pertinent for lymphomas and lung cancer, because RA patients already have an increased risk of these cancers compared to the general population,^8–10 likely related to RA disease activity and chronic immune stimulation.

To date, studies examining the association between TNFis and cancer risk in patients with RA have been mixed. One of the earliest studies was a meta-analysis of nine randomized, placebo-controlled trials of infliximab and adalimumab,⁷ which reported significantly higher overall cancer risk in the treated patients [odds ratio (OR) 3.3, 95%CI 1.2–9.1], with an apparent excess of lymphoma cases. Notably, however, trial participants were only followed for one year or less, and there were only 29 cancer events. Two additional meta-analyses of randomized trials reported no association between TNFi use and overall malignancy.^11,12 However, the first of these found a two-fold increased risk for non-melanoma skin cancer (NMSC, relative risk 2.02, 95%CI 1.11–3.95) while the other reported a non-significant increase in lymphoma risk (OR 2.14, 95%CI 0.55–8.38).^11,12 Some observational studies have also continued to observe positive associations between TNFi use and lymphoma, melanoma, or NMSC risk.^13–21

Collectively, these previous studies suggest that TNFi use may be associated with increased risk of some cancers among patients with RA. However, the studies commonly had short follow-up, grouped malignancies together, or were underpowered to examine specific cancer sites or subtypes. Cancer is heterogenous with respect to natural history and etiology. In particular, non-Hodgkin lymphoma (NHL) is comprised of etiologically distinct subtypes, and analyses of NHL overall may miss relevant associations.²² For example, some NHL subtypes (e.g., diffuse large B-cell lymphoma [DLBCL]) are greatly increased in other immunosuppressed populations, e.g., individuals with acquired immunodeficiency syndrome (AIDS).²³

In the present study, we examined the association between TNFi use and risk of malignancy in RA patients over age 65 years in the US. We utilized data from the Surveillance, Epidemiology, and End Results (SEER)-Medicare database, a large population-based resource, which allowed us to systematically assess distinct solid cancers and hematological malignancies.

Methods

This study was considered non-human subjects research by the National Institutes of Health. We used SEER-Medicare data to perform a case-control study of cancer among individuals with RA.²⁴ SEER-Medicare is a linkage between 18 central cancer registries covering approximately 28% of the US population and Medicare, a federally funded health insurance program for Americans aged 65 years and older or with disability.²⁵ Approximately 93% of SEER cases over age 65 link to Medicare claims. Medicare provides hospital benefits (part A) to all enrollees, and most also receive part B benefits for outpatient and physician care. Since 2006, Medicare beneficiaries are eligible to receive medication benefits through part D. The SEER-Medicare database also includes a 5% random sample of cancer-free Medicare beneficiaries from SEER catchment areas.

We identified all SEER cases with a first cancer diagnosis at age 66–99 (2007–2015). Cancers diagnosed only at autopsy or on death certificate were excluded. Controls were selected for each calendar year 2007–2015 from the 5% Medicare sample. They were required to be cancer-free and aged 66–99 on July 1 in the calendar year of selection. Controls were selected in each calendar year they were eligible and could later become a case. We did not include individuals younger than 65, because their eligibility for Medicare due to disability may have been attributable to severe RA, making them unrepresentative of the RA population.

To identify the subset of cases and controls who had RA, Medicare claims were assessed for each beneficiary starting at study entry, which was defined as the later of age 65 or first calendar year of Medicare claims data. All potential study participants were required to have at least one month of Medicare part A, part B, and part D coverage outside a health maintenance organization (HMO) in the time between study entry and 365 days prior to case diagnosis/control selection (coverage in an HMO was excluded because HMOs do not provide itemized claims to Medicare for services).²⁴ Finally, we required that included study participants have an inpatient Medicare claim with an RA diagnosis (International Classification of Diseases [ICD] 9^th and 10^th Revision codes, see Supplemental Table 1) or an RA diagnosis occurring in the outpatient setting from a rheumatologist occurring at least 365 days prior to selection.

Cancer cases were classified using the SEER site recode. Because SEER-Medicare does not allow reporting of cell sizes <12, we did not separately examine cancer sites for which the expected number of TNFi-exposed cases under the null hypothesis (i.e., number of site-specific cases times the proportion of exposed controls) was below this threshold. Instead, sites with too few cases were grouped with similar sites if possible or moved to a “miscellaneous” category. We categorized NHL cases according to SEER-recorded histology and separately evaluated the three most common subtypes (DLBCL, follicular lymphoma, chronic lymphocytic leukemia/small lymphocytic lymphoma [CLL/SLL]). Hodgkin lymphoma was uncommon and was grouped with miscellaneous hematologic malignancies, but because it is a distinct entity we also examined it separately.

Cutaneous NMSC, basal cell carcinoma (BCC), and squamous cell carcinoma (SCC) diagnoses are not captured by SEER cancer registries. We therefore identified such cases using Medicare claims from the cancer-free 5% random sample of controls, requiring both a diagnosis code for the skin cancer and a procedure code indicating removal (Supplemental Table 1).²⁶ First diagnoses in Medicare for SCC and BCC could be identified only during 2012–2015 when specific ICD-9/10 codes differentiating these diagnoses entered the lexicon. We assessed a first observable diagnosis of the less specific diagnosis NMSC using Medicare claims for the entire 2007–2015 period. NMSC, SCC and BCC controls were selected from the full control set as described above. These controls were further required to not have a diagnosis of the respective skin cancer prior to July 1 of the selection year. See Supplemental Figure 1 for flowchart of case and control selection.

Exposure to TNFis (etanercept, infliximab, adalimumab, golimumab, certolizumab) as well as methotrexate, other disease-modifying antirheumatic drugs (DMARDs), and corticosteroids, was ascertained from part B and part D claims beginning at study entry and ending at cancer diagnosis/control selection (see Supplemental Table 1 for National Drug Code (NDC) and Healthcare Common Procedure Coding System [HCPCS] codes). Individuals with a claim for another biologic (ustekinumab, abatacept, tocilizumab, anakinra, rituximab, tofacitinib) prior to diagnosis/selection were excluded. We also used all claims between study entry and diagnosis/selection to assess potential confounding factors (Supplemental Table 2), including smoking indicators, RA severity, and history of NMSC (V code) or a diagnosis of actinic keratosis (both of which are skin cancer risk factors).^2,27 A person was classified as having a positive smoking history based on a claim for chronic obstructive pulmonary disease, smoking cessation, or history of smoking. Markers of severe RA included complications of RA, i.e., an inpatient RA diagnosis code occurring in the first position, or a claim for Sjogren syndrome, interstitial lung disease, Felty syndrome, or vasculitis. Corticosteroid or opioid use, as ascertained by the presence of a part D claim at any time prior to diagnosis/selection, also signified severe disease. Corticosteroid use was classified as a binary variable, whereas opioid use was classified with the following levels: 0, 1–3, or 4+ part D claims.

Logistic regression was used to estimate adjusted odds ratios (aORs) for each TNFi-cancer association, where the referent group were individuals without evidence of TNFi use. Analyses for melanoma, NMSC, BCC and SCC were restricted to White cases and controls because there were few non-White cases. We present odds ratios minimally adjusted for age, sex and race where appropriate, and cancer diagnosis/control selection year (aOR1 in Table 2), but our primary analyses additionally adjusted for smoking history and methotrexate use prior to diagnosis/selection (aOR2). Primary analyses for melanoma, NMSC, BCC, and SCC were further adjusted for history of NMSC and actinic keratosis. Based on prior evidence, we expected associations between TNFi and NMSC, melanoma, NHL, and Hodgkin lymphoma, so we used a p-value of 0.05 to determine significance of the associations for these cancers; for other cancer sites, we used a Bonferroni correction of 0.0022 = 0.05/22, as 22 additional cancer sites were examined. We used a robust variance calculation to account for multiple sampling of controls across calendar years and because some controls later were cases.²⁴ For associations that were statistically significant in the primary models, we tested for an interaction between TNFi and methotrexate use, and we further adjusted associations for markers of severe RA (complications of RA, corticosteroid use, and level of opioid use). We further conducted sensitivity analyses restricted to individuals with a claim for at least one RA medication (TNFi, methotrexate, or another DMARD) to test the robustness of associations in a population under active medical management for RA.

Table 2.

Associations between TNFi exposure and cancer

Controls or cancer site	Total	TNFi exposed		aOR1	95%CI		aOR2	95%CI
		N	%
Controls	30,475	4,931	16.2	Reference
Head and neck	249	41	16.4	0.99	0.70	1.41	0.96	0.67	1.37
Esophagus	90	19	21.1	1.44	0.84	2.46	1.55	0.88	2.73
Stomach	187	24	12.8	0.87	0.56	1.35	0.81	0.52	1.28
Colon	727	107	14.7	1.02	0.82	1.27	0.99	0.79	1.24
Rectosigmoid/rectum	169	27	16.0	1.09	0.71	1.66	1.02	0.66	1.59
Anus and genital organs	138	21	15.1	0.89	0.55	1.45	0.94	0.58	1.54
Liver	129	20	15.5	0.96	0.59	1.58	1.21	0.72	2.03
Biliary tract	137	20	14.6	1.00	0.61	1.62	0.99	0.59	1.66
Pancreas	392	59	15.0	1.02	0.77	1.37	1.00	0.74	1.34
Lung	2,167	361	16.6	1.01	0.89	1.16	0.93	0.81	1.08
Melanoma of skin a	317	73	23.0	1.53	1.15	2.03	1.33	0.99	1.80
Breast b	1,340	194	14.5	0.85	0.72	1.01	0.85	0.71	1.02
Uterus b	239	42	17.6	1.00	0.71	1.42	0.96	0.68	1.37
Ovary b	158	25	15.8	0.98	0.64	1.52	0.85	0.55	1.33
Prostate b	701	90	12.8	0.73	0.56	0.96	0.75	0.57	0.99
Urinary bladder	464	71	15.3	1.01	0.77	1.33	1.02	0.77	1.35
Kidney	297	55	18.5	1.14	0.84	1.55	1.15	0.84	1.59
Thyroid	117	23	19.7	1.07	0.67	1.71	1.16	0.72	1.87
Hodgkin lymphoma and miscellaneous hematological malignancies	166	36	21.7	1.46	1.00	2.12	1.44	0.96	2.17
NHL	783	176	22.5	1.49	1.24	1.79	1.28	1.06	1.56
Diffuse large B-cell lymphoma	370	78	21.1	1.34	1.03	1.74	1.06	0.81	1.39
Follicular lymphoma	92	31	33.7	2.56	1.63	4.02	2.63	1.63	4.24
CLL/SLL	148	27	18.2	1.19	0.78	1.81	1.15	0.74	1.78
Myeloma	162	22	13.6	0.87	0.55	1.37	0.83	0.52	1.31
Acute myeloid leukemia	99	17	17.2	1.13	0.66	1.94	1.16	0.64	2.10
NMSC c	501	105	21.0	1.43	1.16	1.77	1.32	1.06	1.63
BCC d	161	42	26.1	1.71	1.20	2.44	1.67	1.14	2.44
SCC d	150	23	15.3	0.89	0.58	1.39	0.87	0.54	1.40
Miscellaneous	1,027	169	16.5	1.14	0.95	1.37	1.06	0.88	1.28

Abbreviations: aOR, adjusted odds ratio; TNFi, tumor necrosis factor inhibitor; NHL, non-Hodgkin lymphoma; CLL, chronic lymphocytic leukemia; SLL, small lymphocytic leukemia; NMSC, non-melanoma skin cancer; BCC, basal cell carcinoma; SCC, squamous cell carcinoma. Bolded values are statistically significant if p <0.05 for associations with prior evidence or p <0.0022 for associations without prior evidence.

aOR1: Odds ratio is adjusted for selection year, age, and race and sex where appropriate.

aOR2: Odds ratio is adjusted for selection year, age, race and sex where appropriate, methotrexate use, and smoking indicators. Odds ratios for skin cancers are additionally adjusted for history of NMSC and actinic keratosis diagnoses.

^aAnalyses were restricted to White Medicare beneficiaries and included 23,860 controls.

^bAnalyses were restricted to males (prostate cancer cases) or females (breast, uterus, and ovary cancer cases) and, correspondingly, 6,111 male controls or 24,364 female controls.

^cAnalyses were restricted to White Medicare beneficiaries including 23,428 controls. NMSC cases were selected from the 5% random sample (2007–2015). All White Medicare controls were eligible for control selection when a NMSC diagnosis was not apparent in the years up to and including the selection year.

^dBCC and SCC analyses (2012–2015) were restricted to White Medicare beneficiaries and included 12,511 controls for BCC cases and 12,529 controls for SCC cases.

Additional exploratory analyses were performed to further understand primary associations. First, we examined associations separately for the most commonly used medications (infliximab and etanercept). We performed stage-specific analyses for several cancer sites with a large number of cases, to determine whether TNFi use affected how advanced the cancer was at presentation. To understand when TNFi exposure may act to alter cancer risk, we performed latency analyses for cancers sites with significant overall associations by examining associations with timing of first observed TNFi claim (0–1, 1–2, 2–5, or 5+ years before diagnosis/selection). Finally, we performed analyses for skin cancers restricted to individuals with at least one dermatology visit more than 180 days prior to the diagnosis/selection date. Because these analyses were exploratory, we did not focus on statistical significance but on qualitative comparison to the primary associations.

Results

The study included 10,263 SEER cancer cases, 501 NMSC cases, 161 BCC cases, 150 SCC cases and 30,475 controls with RA (Table 1). Because we selected all eligible controls in every calendar year, controls were selected a median of three times over different calendar years. Compared to controls, a lower proportion of SEER cases were female (69.4% versus 79.9%), and a similar proportion were White (81.7% versus 78.3%). Only 64.0% and 57.1% of SCC and BCC cases, respectively, were female. A large proportion of the study population had evidence of a history of smoking (range: 49.1–56.0% across the case/control groups). A low proportion of cases and controls had indicators of RA complications (12.4–18.0%), although a large fraction had prescriptions for corticosteroids (55.2–65.2%) or opioids (63.5–73.9%). Approximately 40% of cases and controls had a claim for methotrexate, 29.3–36.0% had a claim for another DMARD, and 55.5–68.3% received any RA medication prior to diagnosis/selection. SEER case cancer diagnoses were classified into 26 specific sites or groupings (Table 2).

Table 1.

Demographic and clinical characteristics of cancer cases and controls with rheumatoid arthritis

Characteristic	SEER Cases n = 10,263		NMSC n=501		SCC n=150		BCC n=161		Controls n = 30,475
	N	%	N	%	N	%	N	%	N	%
Age at selection/diagnosis (years)
65–69	1,369	13.3	78	15.6	20	13.3	29	18.0	4,536	14.9
70–74	2,783	27.1	122	24.4	36	24.0	48	29.8	8,223	27.0
75–79	2,509	24.4	96	19.2	37	24.7	27	16.8	6,861	22.5
80–84	1922	18.7	102	20.4	29	19.3	28	17.4	5,571	18.3
85+	1680	16.4	103	20.6	28	18.7	29	18.0	5,284	17.3
Sex
Male	3,140	30.6	144	28.7	54	36.0	69	42.9	6,111	20.1
Female	7,123	69.4	357	71.3	96	64.0	92	57.1	24,364	79.9
Race
White	8,389	81.7	501	100.0	150	100.0	161	100.0	23,860	78.3
Black	1,136	11.1	.	.	.	.	.	.	3,389	11.1
Other	738	7.2	.	.	.	.	.	.	3,226	10.6
Year of diagnosis/selection
2007–2009	2,878	28.0	212	42.3	.	.	.	.	8,080	26.5
2010–2012	3,268	31.8	188	37.5	72	48.0	70	43.5	9,684	31.8
2013–2015	4,117	40.1	101	20.2	78	52.0	91	56.5	12,711	41.7
Smoking
No	4,152	40.5	255	50.9	66	44.0	81	50.3	15,360	50.4
Yes	6,111	59.5	246	49.1	84	56.0	80	49.7	15,115	49.6
Complications of rheumatoid arthritis a
No	8,787	85.6	421	84.0	123	82.0	141	87.6	26,248	86.1
Yes	1,476	14.4	80	16.0	27	18.0	20	12.4	4,227	13.9
Opioid use
None	2,988	29.1	183	36.5	43	28.7	42	26.1	9,639	31.6
1–3 prescriptions	2,802	27.3	169	33.7	49	32.7	53	32.9	7,902	25.9
4+ prescriptions	4,473	43.6	149	29.7	58	38.7	66	41.0	12,934	42.4
Any treatment b
No	4,237	41.3	204	40.7	54	36.0	51	31.7	13,564	44.5
Yes	6,026	58.7	297	59.3	96	64.0	110	68.3	16,911	55.5
Corticosteroid use
No	4,065	39.6	232	44.8	48	32.0	56	34.8	223	44.5
Yes	6,198	60.4	286	55.2	102	68.0	105	65.2	278	55.5
Methotrexate use
No	6,145	59.9	310	61.9	89	59.3	90	55.9	19,100	62.7
Yes	4,118	40.1	191	38.1	61	40.7	71	44.1	11,375	37.3
Other DMARD use c
No	7,130	69.5	349	69.7	92	61.3	103	64.0	21,546	70.7
Yes	3,133	30.5	152	30.3	58	38.7	58	36.0	8,929	29.3
TNFi use
No	8,571	83.5	396	79.0	127	84.7	119	73.9	25,544	83.8
Yes	1,692	16.5	105	21.0	23	15.3	42	26.1	4,931	16.2
TNFi medication
Etanercept	508	5.0	30	6.0	NR	NR	14	8.7	1,681	5.5
Infliximab	907	8.8	64	12.8	13	8.7	21	13.0	2,384	7.8
Adalimumab	368	3.6	16	3.2	NR	NR	NR	NR	1,199	3.9
Golimumab	52	0.5	NR	NR	NR	NR	NR	NR	130	0.4
Certolizumab	65	0.6	NR	NR	NR	NR	NR	NR	226	0.7

Abbreviations: NMSC, non-melanoma skin cancer; BCC, basal cell carcinoma; SCC, squamous cell carcinoma; TNFi, tumor necrosis factor inhibitor; MTX, methotrexate; DMARD, disease modifying antirheumatic drug; NR, not reported because cell size < 12.

^aA person is classified as having complications of RA if an RA diagnosis occurs in the first position of a hospitalization claims or if there is a diagnosis for any of the following conditions prior to cancer diagnosis/control selection: Sjogren syndrome, interstitial lung disease, Felty syndrome, or vasculitis. See Supplemental Table 2 for diagnosis codes.

^bAny treatment is defined as methotrexate, other DMARD, or TNFi use.

^cOther DMARDs include hydroxychloroquine, leflunomide, sulfasalazine, azathioprine, cyclophosphamide, cyclosporine, penicillamine, and minocycline. See Supplemental Table 1.

TNFi exposure prior to diagnosis/selection was observed in 16.5% of SEER cancer cases, 21.0% of NMSC cases generally, 15.3% of SCC cases, 26.1% of BCC cases, and 16.2% of controls (Table 1). TNFi use varied considerably by cancer site from 12.8% for stomach cancer to 33.7% for follicular lymphoma (Table 2). In the primary analyses, TNFi exposure was associated with risk of NHL overall (aOR2 1.28, 95%CI 1.06–1.56), follicular lymphoma (2.63, 1.63–4.24), and NMSC overall (1.32, 1.06–1.63). Exposure was not associated with Hodgkin lymphoma and miscellaneous hematologic malignancies (Table 2) or Hodgkin lymphoma separately (N=51 cases, aOR2 1.32, 95%CI 0.67–2.61). Using a Bonferroni correction, we observed borderline significant inverse associations for breast cancer (aOR2 0.85, 95%CI 0.71–1.02) and prostate cancer (0.75, 0.57–0.99), and borderline positive associations for BCC (1.67, 1.14–2.44) and melanoma (1.33, 0.99–1.80).

Results were similar in a minimally adjusted model (aOR1, Table 2). The interactions between TNFi and methotrexate use were not significant for NHL overall (p=0.33), follicular lymphoma (p=0.61), or NMSC overall (p=0.99). All statistically significant associations in Table 2 remained significant and similar when further adjusted for markers of severe RA (Supplemental Table 3). The association between TNFi use and NMSC overall (aOR 1.44, 95%CI 1.12–1.85) was slightly stronger, and the association with BCC was unchanged (1.67, 1.08–2.58), when analyses were restricted to individuals with at least one dermatology visit more than 180 days prior to diagnosis/selection (Supplemental Table 4). When we restricted to individuals observed to receive at least one RA medication, results were similar, although slightly attenuated (Supplemental Table 5).

We also examined associations between infliximab or etanercept and individual cancer sites, and a few qualitative differences were suggested (Table 3). Notably, etanercept use was associated with greater than two-fold risk of esophageal cancer (aOR 2.39, 95%CI 1.17–4.86), and more than a 40% reduction in prostate cancer risk (0.55, 0.34–0.89). Infliximab and etanercept were both associated with follicular lymphoma. There were no clear patterns with respect to the associations between TNFi exposure and stage of cancer diagnosis (Supplemental Table 6).

Table 3.

Associations between exposure to TNFi, infliximab, or etanercept and cancer

Controls or cancer site	Total	TNFi exposed		aOR	95%CI		Infliximab exposed		aOR	95%CI		Etanercept exposed		aOR	95%CI
		N	%				N	%				N	%
Controls	30,475	4,931	16.2	Ref			2,384	7.8	Ref			1,690	5.5	Ref
Head and neck	249	41	16.4	0.96	0.67	1.37	26	10.4	1.35	0.88	2.07	NR	NR	0.67	0.35	1.30
Esophagus	90	19	21.1	1.55	0.88	2.73	NR	NR	1.36	0.65	2.84	NR	NR	2.39	1.17	4.86
Stomach	187	24	12.8	0.81	0.52	1.28	13	7.0	0.88	0.49	1.57	NR	NR	0.52	0.21	1.28
Colon	727	107	14.7	0.99	0.79	1.24	61	8.4	1.11	0.83	1.48	29	4.0	0.82	0.55	1.22
Rectosigmoid/rectum	169	27	16.0	1.02	0.66	1.59	NR	NR	0.63	0.31	1.28	NR	NR	1.27	0.67	2.40
Anus and genital organs	138	21	15.1	0.94	0.58	1.54	NR	NR	1.12	0.59	2.13	NR	NR	0.54	0.20	1.46
Liver	129	20	15.5	1.21	0.72	2.03	NR	NR	1.31	0.66	2.62	NR	NR	1.17	0.54	2.52
Biliary tract	137	20	14.6	0.99	0.59	1.66	13	9.5	1.42	0.78	2.58	NR	NR	1.02	0.46	2.27
Pancreas	392	59	15.0	1.00	0.74	1.34	27	6.9	0.91	0.60	1.38	18	4.6	0.91	0.55	1.49
Lung	2,167	361	16.6	0.93	0.81	1.08	192	8.9	1.00	0.83	1.21	104	4.8	0.80	0.63	1.02
Melanoma of skin a	317	73	23.0	1.33	0.99	1.80	44	13.9	1.50	1.05	2.16	21	6.6	1.11	0.69	1.78
Breast b	1,340	194	14.5	0.85	0.71	1.02	104	7.8	0.99	0.79	1.25	61	4.6	0.80	0.60	1.07
Uterus b	239	42	17.6	0.96	0.68	1.37	25	10.5	1.32	0.86	2.04	NR	NR	0.50	0.24	1.03
Ovary b	158	25	15.8	0.85	0.55	1.33	NR	NR	0.75	0.40	1.42	14	8.9	1.59	0.89	2.83
Prostate b	701	90	12.8	0.75	0.57	0.99	50	7.1	0.86	0.60	1.24	23	3.3	0.55	0.34	0.89
Urinary bladder	464	71	15.3	1.02	0.77	1.35	37	8.0	1.06	0.74	1.52	20	4.3	0.83	0.51	1.34
Kidney	297	55	18.5	1.15	0.84	1.59	32	10.8	1.39	0.94	2.06	19	6.4	1.13	0.69	1.86
Thyroid	117	23	19.7	1.16	0.72	1.87	NR	NR	0.84	0.40	1.76	NR	NR	0.95	0.43	2.11
Hodgkin lymphoma and miscellaneous hematological malignancies	166	36	21.7	1.44	0.96	2.17	19	11.4	1.49	0.91	2.45	NR	NR	1.19	0.62	2.28
NHL	783	176	22.5	1.28	1.06	1.56	97	12.4	1.35	1.06	1.72	60	7.7	1.26	0.94	1.69
Diffuse large B-cell lymphoma	370	78	21.1	1.06	0.81	1.39	41	11.1	1.08	0.77	1.53	26	7.0	1.05	0.69	1.62
Follicular lymphoma	92	31	33.7	2.63	1.63	4.24	18	19.6	2.64	1.51	4.62	NR	NR	2.00	1.01	3.98
CLL/SLL	148	27	18.2	1.15	0.74	1.78	17	11.5	1.44	0.85	2.47	NR	NR	1.27	0.66	2.47
Myeloma	162	22	13.6	0.83	0.52	1.31	NR	NR	0.68	0.33	1.39	NR	NR	0.63	0.28	1.45
Acute myeloid leukemia	99	17	17.2	1.16	0.64	2.10	NR	NR	1.34	0.66	2.74	NR	NR	0.58	0.18	1.88
NMSC a	501	105	21.0	1.32	1.06	1.63	64	12.8	1.37	1.06	1.77	30	6.0	1.22	0.85	1.75
BCC a	161	42	26.1	1.67	1.14	2.44	21	13.0	1.54	0.93	2.56	14	8.7	1.52	0.87	2.64
SCC a	150	23	15.3	0.87	0.54	1.40	13	8.7	1.02	0.56	1.87	NR	NR	1.04	0.51	2.12
Miscellaneous	1,027	169	16.5	1.06	0.88	1.28	89	8.7	1.08	0.85	1.38	49	4.8	0.92	0.67	1.26

Abbreviations: aOR, adjusted odds ratio; TNFi, tumor necrosis factor inhibitor; NHL, non-Hodgkin lymphoma; CLL, chronic lymphocytic leukemia; SLL, small lymphocytic leukemia; NMSC, non-melanoma skin cancer; BCC, basal cell carcinoma; SCC, squamous cell carcinoma; NR, not reported because cell size < 12. Bolded values signify associations where p <0.05.

aOR: Odds ratio is adjusted for selection year, age, race and sex where appropriate, methotrexate use, and smoking indicators.

^aAnalyses were restricted to White Medicare beneficiaries, See Table 2 notes for details of control selection. Odds ratios for skin cancers are additionally adjusted for history of NMSC and actinic keratosis diagnoses appearing in claims prior to diagnosis/selection.

^bAnalyses were restricted to males (prostate cancer cases) or females (breast, uterus, and ovary cancer cases) and, correspondingly, 6,111 male controls or 24,364 female controls.

In latency analyses (Table 4), the association between TNFi use and follicular lymphoma was strongest when the first TNFi claim was observed 1–2 years before diagnosis/selection (aOR 4.34, 95%CI 1.94–9.71), but the aOR was elevated for all periods at least one year prior to diagnosis/selection. There was an increased risk for NHL regardless of when the first TNFi claim was observed. The association between TNFi use and NMSC was elevated at all time points except in the 2–5 years prior to diagnosis/selection.

Table 4.

Latency analyses for associations between TNFi exposure and selected cancers

Cancer site	Cases		Controls		aOR	95%CI		p trend
	N	%	N	%
NHL
Never exposed	607	77.5	25,544	83.8	-	-	-
Ever exposed	176	22.5	4,931	16.2	1.49	1.24	1.79
Exposure according to earliest claim								0.53
0–1 year prior	20	2.6	555	1.8	1.54	0.97	2.44
1–2 year prior	27	3.5	776	2.6	1.51	1.01	2.26
2–5 years prior	75	9.6	1892	6.2	1.71	1.33	2.21
5+ years prior	54	6.9	1708	5.6	1.32	0.98	1.78
Follicular lymphoma
Never exposed	61	66.3	25,544	83.8	-	-	-
Ever exposed	31	33.7	4,931	16.2	2.56	1.63	4.02
Exposure according to earliest claim								0.93
0–1 year prior	NR	NR	555	1.8	1.58	0.39	6.37
1–2 year prior	NR	NR	776	2.6	4.34	1.94	9.71
2–5 years prior	NR	NR	1892	6.2	2.19	1.09	4.38
5+ years prior	NR	NR	1708	5.6	2.62	1.36	5.06
NMSC
Never exposed	396	78.9	19,541	86.3	-	-	-
Ever exposed	106	21.1	3,905	16.7	1.43	1.16	1.77
Exposure according to earliest claim								0.08
0–1 year prior	21	4.2	424	1.8	2.19	1.38	3.48
1–2 year prior	21	4.2	590	2.5	1.90	1.19	3.01
2–5 years prior	29	5.8	1,489	6.4	1.02	0.70	1.50
5+ years prior	35	7.0	1,402	6.0	1.40	0.99	1.97

Abbreviations: aOR, adjusted odds ratio; TNFi, tumor necrosis factor inhibitor; NHL, non-Hodgkin lymphoma; NMSC, non-melanoma skin cancer; NR: not reported because cell size < 12

aOR: Odds ratio is adjusted for selection year, age, and race and sex where appropriate. Analyses for NMSC were restricted to White Medicare beneficiaries

Discussion

In this study of TNFis and cancer risk among older adults with RA, TNFi exposure was significantly associated with an increased risk for follicular lymphoma and the broader groupings of NHL and NMSC. Notably, we did not observe significantly increased risk associated with TNFi exposure for a range of other common cancers. Because TNFi exposure is associated with an increased risk of infection and the reactivation of latent infections such as tuberculosis, there have been concerns that TNFis have immunosuppressive effects.^5,6 However, the role of TNF in cancer is complex, and TNFis may have both cancer promoting and limiting effects that are separate from their effects on immunity and depend on the specific cancer.²⁸

Although many studies have reported positive associations between TNFi use and lymphoma risk^{8,10,12,14,29,30}, the evidence is still mixed with several studies reporting no association ^15,31,32 Previous studies were generally unable to examine specific lymphoma subtypes. This could obscure results, because Hodgkin lymphoma and NHL are distinct, and NHL itself comprises a large number of subtypes which have unique etiologies.²² The significant association with NHL in our study seemed to be largely driven by follicular lymphoma. Unlike DLBCL and certain other NHL subtypes, follicular lymphoma is not elevated in immunosuppressed populations such as persons with AIDS or transplant recipients,^23,33 and to the best of our knowledge, RA itself is not associated with follicular lymphoma as it is with DLBCL.³⁴ Furthermore, while DLBCL in immunosuppressed populations is often associated with Epstein-Barr virus (EBV), very few follicular lymphomas are EBV-positive.³⁵

The association between TNFi exposure and follicular lymphoma is novel. We are aware of only one prior study with data on follicular lymphoma, which evaluated a cohort of 11,931 TNFi-treated RA patients in the United Kingdom.³¹ Data presented in Table 2 of that 2015 study support an increased risk of follicular lymphoma compared to unexposed patients (incidence rate ratio 3.7), but there were only 19 cases in total and the association was not significant (chi-square p-value = 0.17 based on our calculation). Of interest, a substantial proportion of follicular lymphomas harbor a mutation in TNFRSF14 (a gene in the TNF receptor superfamily) that results in reduced function.³⁶ The mechanistic implications of this mutation are uncertain, but one possibility is that TNFis may promote the transformation of premalignant lymphocytes to cause follicular lymphoma through a similar pathway. Follicular lymphoma cells represent the pathological counterpart of normal germinal center B-cells and are characterized by a t(14;18) translocation between the immunoglobulin heavy chain gene and the BCL2 oncogene.³⁶ Although >70% of healthy individuals carry low levels of t(14;18)+ B-cells in the blood or lymphoid tissues, the vast majority will never develop follicular lymphoma, indicating that BCL2 deregulation alone is insufficient to transform B lymphocytes.

Hodgkin lymphoma and miscellaneous hematological malignancies were relatively uncommon in our study, and we observed only a borderline association between TNFi exposure and this malignancy. It would have been preferable to analyze Hodgkin lymphoma and the more common subtypes separately, but this study was not statistically powered for these analyses. There have been case reports of hepatosplenic T-cell lymphoma arising in children and adolescents treated for IBD with TNFis, resulting in a Food and Drug Administration black-box warning.^37,38 We would have liked to examine this aggressive NHL subtype, but there were no cases in our study, which is not surprising given its rarity and demographic profile.³⁹

Chronic ultraviolet radiation exposure is the major cause of melanoma and NMSCs.⁴⁰ Skin cancers manifest a high tumor mutational burden caused by ultraviolet radiation, and the elevated risk in immunosuppressed populations is thought to be due to loss of immune control directed against tumor neoantigens. We observed a borderline association between TNFis and melanoma and found a significant association with NMSC generally, as has been reported by others.¹³ However, we did not replicate a previously reported association with SCC,¹⁹ and observed a moderately strong (albeit statistically insignificant) association with BCC. One possible explanation for these associations with skin cancers is that patients receiving TNFis may receive more frequent skin cancer screening than other RA patients, leading to increased detection. However, when we restricted analyses to individuals with at least one dermatology visit more than 180 days prior to diagnosis/selection, the associations with BCC and NMSC overall persisted.

We observed a few additional borderline statistical associations, including inverse associations between TNFi exposure and prostate and breast cancers. TNF is highly expressed in some tumors including breast, prostate, ovarian, and colorectal tumors, and clinical trials have examined the treatment benefits of etanercept for breast and ovarian cancers.^41,42 It therefore may be biologically plausible that TNFis could prevent some cancers from developing. Two prior studies have suggested inverse associations with risk of breast or colorectal cancer.^20,43

We performed latency analyses to understand how the association between TNFis and cancers varied according to when the first TNFi claim between study entry and selection was observed. There was a moderate increase in NHL risk regardless of when the first TNFi exposure occurred. Follicular lymphoma risk was increased starting more than 1 year after the first documented exposure, but the risk appeared greater in the 1–2-year period compared to the other intervals, suggesting that TNFi exposure may have accelerated an already occurring carcinogenic process, e.g., by promoting expansion of t(14;18)+ B-cell clones.

Our study had several strengths. To our knowledge, it is the largest to date to evaluate the associations between TNFis and cancer risk, as can be judged by comparing the number of cancer cases in our study with those in prior cohort studies.^{13,14,17,18,20,21,29,31,32,43} The large number of cases and comprehensive data in cancer registries allowed us to examine distinct cancer sites and subtypes, rather than grouping cancers as solid or hematological malignancies. Although we made multiple comparisons, we employed a conservative Bonferroni correction to mitigate identifying spurious associations in our primary analyses. Finally, we performed several exploratory analyses to further examine the associations which we identified.

A limitation to our study is that it only included RA patients older than age 65. However, because the incidence of most cancers increases strongly with age, our population-based sample included an important group in which to assess the effects of TNFis. Perhaps of greater salience, we lacked life-course information on RA treatment, and because RA has a median age of onset of 30–60 years, we could not assess or account for earlier therapies. Indeed, the median duration of Medicare claims data prior to cancer diagnosis among cases was only 6.7 years (interquartile range 6.1–7.3). Because of this limited window, we did not attempt a new user analysis, which would have allowed an assessment of early and cumulative effects of TNFi exposure.^44,45 Finally, we lacked detailed clinical information on RA disease severity. Although we adjusted for proxies for severe disease, some associations that we observed could have been confounded by disease severity. However, this issue may not be a major limitation, because we are not aware that severe RA by itself is associated with risk of skin cancers or follicular lymphoma, although there are reports of association with DLBCL.^34,46,47 Finally, we had concerns that the TNFi-follicular lymphoma association could in part be attributable to detection bias, because some cases of follicular lymphoma have an indolent presentation and thus might be differentially diagnosed.⁴⁸ However, CLL/SLL is also indolent, and there was no association with CLL/SLL.

A few limitations were specific to NMSC analyses. Because NMSCs are not captured in US cancer registries, we used a claims-based algorithm to identify cases.⁴⁹ However, because the algorithm incorporates a surgical procedure, we expect that the algorithm is very specific which would result in minimal bias in the measured associations. Also, the identified NMSCs are unlikely to have been a person’s first NMSC, since we could not identify these prior to age 65. The inability to capture and adjust for all prior skin cancer or precursor lesions might have biased the NMSC associations towards the null, if providers were reluctant to prescribe TNFis to individuals with such a history, because previous skin cancers and precursor lesions are strong risk factors for future NMSC. We were also unable to adjust for cumulative ultraviolet radiation exposure.

Reassuringly, risk of most cancers was not elevated among TNFi-exposed persons. We also did not observe an increased risk of cancers that are increased in immunosuppressed individuals such as DLBCL and human papillomavirus-associated cancers, which argues against strong immunosuppressive mechanisms. Furthermore, the associations between TNFi use and NMSC were modest. However, given the consistency with other studies, individuals initiating a TNFi should still receive regular skin cancer screenings and engage in sun protection measures, particularly if they are otherwise at elevated risk.

In conclusion, our study provides evidence for a previously unsuspected association between TNFi exposure and follicular lymphoma, which may drive the associations with lymphoma reported in other studies. Although these findings may have an immune explanation, the lack of strong immunosuppression in RA patients treated with TNFis and the pattern of association with cancer suggest that other mechanisms could underlie this increased risk.

Abbreviations:

aOR

adjusted odds ratio

CI

confidence interval

TNF

Tumor necrosis factor

TNFis

Tumor necrosis factor inhibitors

AIDS

acquired immunodeficiency syndrome

EBV

Epstein-Barr virus

RA

rheumatoid arthritis

SEER

Surveillance, Epidemiology, and End Results

ICD

International Classification of Diseases

NHL

non-Hodgkin lymphoma

NMSC

non-melanoma skin cancer

BCC

basal cell carcinoma

SCC

squamous cell carcinoma