Tumor necrosis factor-alpha inhibitors and risk of non-Hodgkin lymphoma in a cohort of adults with rheumatologic conditions

Abstract

Based on limited evidence, the U.S. Food and Drug Administration (FDA) issued a black box warning for the use of tumor necrosis factor-alpha inhibitors (TNFIs) and risk of non-Hodgkin lymphoma (NHL). Our objective was to determine the risk of NHL associated with TNFI use by duration and type of anti-TNF agent. We performed a nested case-control study within a retrospective cohort of adults with rheumatologic conditions from a U.S. commercial health insurance database between 2009 and 2015. Use of TNFIs (infliximab, adalimumab, etanercept, golimumab and certolizumab pegol) and conventional-synthetic disease-modifying antirheumatic drugs (csDMARDs) was identified, and conditional logistic regression models were used to estimate adjusted odds ratios (OR) and 95% confidence intervals (CI) for risk of NHL. From a retrospective cohort of 55,446 adult patients, 101 NHL cases and 984 controls matched on age, gender and rheumatologic indication were included. Compared to controls, NHL cases had greater TNFI use (33% versus 20%) but were similar in csDMARD use (70% versus 71%). TNFI ever-use was associated with nearly two-fold increased risk of NHL (OR=1.93; 95% CI: 1.16-3.20) with suggestion of increasing risk with duration (P-trend=0.05). TNF fusion protein(etanercept) was associated with increased NHL risk (OR=2.73; 95% CI: 1.40-5.33), whereas risk with anti-TNF monoclonal antibodies was not statistically significant (OR=1.77; 95% CI: 0.87-3.58). In sensitivity analyses evaluating confounding by rheumatologic disease severity, channeling bias was not likely to account for our results. Our findings support the FDA black box warning for NHL. Continued surveillance and awareness of this rare but serious adverse outcome are warranted with new TNFIs and biosimilar products forthcoming.

INTRODUCTION

The introduction of tumor necrosis factor-α inhibitors (TNFI) represents a significant advancement in the disease-modifying treatment of rheumatologic conditions, including in rheumatoid arthritis (RA), psoriatic arthritis (PsA) and ankylosing spondylitis (AS). However, risk of serious adverse effects such as malignancies remains an important concern.¹ Tumor necrosis factors have pleiotropic roles in angiogenesis, cell proliferation, differentiation and apoptosis, leading to potential pro-malignancy as well as anti-malignancy roles.^{2, 3} The U.S. Food and Drug Administration (FDA) issued a black box warning for the use of TNFIs and risk of non-Hodgkin lymphoma (NHL) in 2009⁴ and alerted the public regarding case reports of hepatosplenic T-cell lymphoma, a rare NHL subtype, in 2011.⁵

Several observational studies^6–19 and meta-analyses^{20, 21} have evaluated the risk of lymphoma with TNFIs. While findings from these studies are inconsistent, recent analyses suggest no excess NHL risk related to TNFI use.^{8, 9, 14, 18, 20} In a 2011 meta-analysis of observational and prospective registry studies of patients with RA, PsA or AS, no significantly elevated risk was found (pooled risk estimate 1.11, 95% confidence interval (CI) 0.70-1.51).²⁰ More recently, investigators using the British Society for Rheumatology Biologics Registrar for Rheumatoid Arthritis (BSRBR-RA) concluded that their study “ruled out” the risk of lymphomas in patients with RA for up to 5 years after treatment initiation.¹⁸ While these past observational and registry-based studies documented the experience of patients with rheumatologic disease over several years, many were limited by self-reported medication use and lack of information on newer TNFI agents.

Patient characteristics and treatment patterns with TNFIs for rheumatologic conditions also vary geographically and over time.^22–24 Compared to the United Kingdom and other European countries where there may be more restrictive coverage for TNFIs, U.S. patients initiating TNFIs tend to have lower disease severity.²³ Early use of TNFIs, both as monotherapy and in combination with other conventional-synthetic disease-modifying antirheumatic drugs (csDMARDs) is also increasing in the U.S.²⁵ Therefore, the goal of this study was to evaluate the risk of NHL associated with TNFI use in an insured U.S. population treated for RA, PsA or AS in more recent years (2010-2014). We also assessed the impact of duration of TNFI use, TNFI type (monoclonal antibody versus fusion protein) and history of csDMARD use on NHL risk.

METHODS

We conducted a nested case-control study within a large U.S. health insurance-based retrospective cohort of adults from the Truven Health MarketScan Research Database. This database contains healthcare utilization data including inpatient admissions, outpatient encounters and pharmacy dispensing data for over 90 million commercially insured beneficiaries.²⁶

Study Population

Our cohort included patients aged 18 or older with a prevalent rheumatologic condition with an approved indication for TNFI therapy (RA, PsA or AS) between January 1, 2010 and December 31, 2014 followed by active treatment for their condition. Rheumatologic conditions were defined using International Classification of Disease-9^th Clinical Modification (ICD-9-CM) codes based on a previously validated algorithm.²⁷ Qualifying patients with RA, PsA or AS were required to have: (i) 1+ hospital discharge diagnoses or (ii) 2+ outpatient diagnoses within a 6-month period with at least one diagnosis coming from a visit to a rheumatologist. Active treatment status was ascertained with the presence of two or more procedural codes for medication administration or pharmacy dispensing claims after RA/PsA/AS diagnosis for the following: (i) prescription non-steroidal anti-inflammatory drugs (NSAIDs); (ii) oral, intravenous or intra-articular corticosteroids; (iii) csDMARDs and/or (iv) biologic DMARDs including TNFIs. Since our study included those with prevalent rheumatologic conditions, the cohort entry date for qualifying patients was defined as the date of their first documented diagnosis. Our cohort was limited to patients with at least 12 months of continuous enrollment (maximum permissible gap of 35 of 365 days) prior to and following cohort entry, without prevalent cancer, HIV/AIDS or history of hematopoietic stem cell transplantation. Patients were followed from cohort entry until the first of any cancer diagnosis, disenrollment from the health plan (defined as a membership lapse of 92+ days) or the end of the study (December 31, 2015). Data collection spanned from 12 months prior to cohort entry through the end of follow up. The Institutional Review Board of the University of Illinois at Chicago approved this study.

Case Ascertainment and Control Selection

Cases of subsequent NHL were defined as one or more inpatient, or two or more outpatient claims at least 30 days apart with ICD-9-CM codes for NHL (200.0-200.8, 202.0-202.0, 202.7-202.9) per an algorithm for administrative claims data in patients with indications for TNFI therapy.^{28, 29} In our nested case-control design, we selected controls by incidence density sampling up to ten controls from the retrospective cohort who were at risk of the outcome of interest (i.e., NHL) at the time of case occurrence (reference time) and survived at least as long as the index case (defined as time since cohort entry).³⁰ Incidence density-sampled controls were matched to their index case based on attained age (5-year interval), gender, and rheumatologic indication. This approach was chosen because of the time-varying nature of medication use, the changes in prescribing practices over time and the required assumptions of lag time and latency reasonable for inferences made about cancer risk.³¹ Also, under these risk set-based sampling conditions, this method produces unbiased approximations of the hazard ratio.³²

Exposure Classification

Exposures of interest included infliximab, etanercept, adalimumab, golimumab and certolizumab pegol. Following their cohort entry date, patients were defined as a user of a medication of interest if they had 2+ pharmacy dispensings (or outpatient administrations) during any 6-month period that spanned from 12 months prior to cohort entry through end of follow up (excluding the 6 months prior to the case-control reference date, or lag period). While patients were required to meet these criteria for a single medication of interest (e.g., 2+ adalimumab dispensings), we further grouped this exposure as ‘any TNFI’ in our main analysis and performed stratified analyses by TNFI type, defined as either TNF fusion protein (etanercept) or anti-TNF monoclonal antibody, and by tertiles of TNFI duration of use (<1.5, 1.5-2.5, >2.5 years).

Covariates

We also collected information on pharmacy dispensings (past and current) for prescription non-steroidal anti-inflammatory drugs (NSAIDs), oral corticosteroids and csDMARDs (i.e., methotrexate, hydroxychloroquine, sulfasalazine, and leflunomide). We collected information on a priori potential confounders including patient demographics, enrollment and inpatient and outpatient diagnoses.

Statistical analysis

The association between TNFI use and subsequent NHL was calculated as odds ratios (OR) with 95% confidence intervals (95% CI) using a multivariable conditional logistic regression model. Since cases and controls were matched on their rheumatologic condition, age, gender, and time since cohort entry, the crude OR is adjusted for these factors. Additionally, we adjusted the OR for a set of clinically relevant and empirically significant confounders (Charlson Comorbidity Index³³ (CCI) scores, use of oral corticosteroids, prescription NSAIDs, and csDMARDs). Findings were considered to be statistically significant at an alpha level of 0.05. While we did not have a direct measure of disease severity in this cohort of patients with rheumatologic conditions, we explored the potential bias of differential high disease severity (disease activity scores [DAS]³⁴ of 5.1 or greater) that are an indication for “step-up” therapy using biologic DMARDs.³⁵ In a deterministic bias analysis,^36–39 we evaluated the degree of unmeasured confounding due to high rheumatologic disease severity among TNFI users that would be required to entirely explain our findings using simulated conditional logit models. Also, given that the rheumatologic indication for TNFI therapy for >80% of the cases and controls was rheumatoid arthritis we performed a subgroup analysis restricted to rheumatoid arthritis patients only.

RESULTS

A total of 101 cases were matched to 984 controls (Table 1). Cases and controls were similar in distribution of age, gender and qualifying rheumatologic condition. Compared to controls, cases had higher CCI scores, lower use of prescription NSAIDs, and greater use of concurrent oral corticosteroids during follow-up. Etanercept was the most commonly used TNFI, followed by infliximab. Ever use of TNFIs was greater among cases (32.7%) than controls (20.2%). TNFI users (n=232) were younger and more likely to have AS or PsA compared to nonusers (n=853) (Table 2). Use of csDMARDs was more prevalent among TNFI users in general, although use of methotrexate was higher and hydroxychloroquine was lower than TNFI nonusers. Oral corticosteroids and NSAID use during follow-up was not significantly different according to TNFI use, and CCI scores were also similar between TNFI users and nonusers.

Table 1.

Characteristics of study subjects by case-control status

	Controls (n=984)		Cases (n=101)
	n	(%)	n	(%)	P *
Gender
Female	662	(67.3)	68	(67.3)	0.99
Male	322	(32.7)	33	(32.7)
Age, years
Median (interquartile range)	58	(51 – 67)	58	(53 – 68)	0.97
30-34	18	(1.8)	2	(2.0)	0.99
35-39	38	(3.9)	4	(4.0)
40-44	50	(5.1)	5	(5.0)
45-49	90	(9.1)	9	(8.9)
50-54	160	(16.3)	16	(15.8)
55-59	180	(18.3)	19	(18.8)
60-64	170	(17.3)	17	(16.8)
65-69	70	(7.1)	7	(6.9)
70-74	95	(9.7)	10	(9.9)
75-79	70	(7.1)	7	(6.9)
80-84	27	(2.7)	3	(3.0)
85-89	16	(1.6)	2	(2.0)
Rheumatologic indication for TNFI therapy
Rheumatoid arthritis	860	(87.4)	87	(86.1)	0.92
Psoriatic arthritis	83	(8.4)	9	(8.9)
Ankylosing spondylitis	41	(4.2)	5	(5.0)
Comorbid conditions
Sjogren’s syndrome	23	(2.3)	1	(1.0)	0.72
Systemic lupus erythematosus	31	(3.2)	2	(2.0)	0.76
Celiac disease	2	(0.2)	0	(0.0)	0.82
Charlson comorbidity index at baseline
0	575	(58.4)	50	(49.5)	0.03
1	262	(26.6)	26	(25.7)
2+	129	(13.1)	23	(22.8)
Ever use of medications in follow up
Prescription NSAIDs	612	(62.2)	50	(49.5)	0.01
Oral corticosteroids	650	(66.1)	78	(77.2)	0.02
Any conventional DMARDs	703	(71.4)	71	(70.3)	0.81
Hydroxychloroquine	322	(32.7)	29	(28.7)	0.41
Sulfasalazine	115	(11.7)	8	(7.9)	0.26
Methotrexate	474	(48.2)	52	(51.5)	0.53
Leflunomide	81	(8.2)	8	(7.9)	0.91
Any TNFI	199	(20.2)	33	(32.7)	<0.01
Etanercept	104	(10.6)	16	(15.8)	0.11
Infliximab	42	(4.3)	5	(5.0)	0.75
Adalimumab	85	(8.6)	14	(13.9)	0.08
Golimumab	12	(1.2)	2	(2.0)	0.38
Certolizumab pegol	9	(0.9)	1	(1.0)	0.94

^*To compare differences by case-control status we used chi-square test for categorical variables (Fisher’s exact test with cells <5) and Wilcoxon rank-sum test for medians

Table 2.

Characteristics of study subjects by ever use of TNFIs

	TNFI nonusers (n=853)		TNFI users (n=232)
	n	(%)	n	(%)	P *
Gender
Female	585	(68.6)	145	(62.5)	0.08
Male	268	(31.4)	87	(37.5)
Age, years
Median (interquartile range)	59	(52 – 69)	55	(50 – 61)	<0.01
30-34	14	(1.6)	6	(2.6)	<0.01
35-39	32	(3.8)	10	(4.3)
40-44	38	(4.5)	17	(7.3)
45-49	79	(9.3)	20	(8.6)
50-54	128	(15.0)	48	(20.7)
55-59	144	(16.9)	55	(23.7)
60-64	147	(17.2)	40	(17.2)
65-69	65	(7.6)	12	(5.2)
70-74	92	(10.8)	13	(5.6)
75-79	71	(8.3)	6	(2.6)
80-84	26	(3.0)	4	(1.7)
85-89	17	(2.0)	1	(0.4)
Rheumatologic indication for TNFI therapy
Rheumatoid arthritis	781	(91.6)	166	(71.6)	<0.01
Psoriatic arthritis	46	(5.4)	46	(19.8)
Ankylosing spondylitis	26	(3.0)	20	(8.6)
Comorbid conditions
Sjogren’s syndrome	23	(2.7)	1	(0.4)	0.04
Systemic lupus erythematosus	30	(3.5)	3	(1.3)	0.13
Celiac disease	2	(0.2)	0	(0.0)	0.62
Charlson comorbidity index at baseline
0	484	(56.7)	141	(60.8)	0.35
1	231	(27.1)	57	(24.6)
2+	125	(14.7)	27	(11.6)
Ever use of other medications in follow up
Prescription NSAIDs	525	(61.5)	137	(59.1)	0.49
Oral corticosteroids	568	(66.6)	160	(69.0)	0.49
Any conventional DMARDs	596	(69.9)	178	(76.7)	0.04
Hydroxychloroquine	303	(35.5)	48	(20.7)	<0.01
Sulfasalazine	93	(10.9)	30	(12.9)	0.39
Methotrexate	374	(43.8)	152	(65.5)	<0.01
Leflunomide	58	(6.8)	31	(13.4)	<0.01

^*To compare differences by ever use of TNFIs we used chi-square test for categorical variables (Fisher’s exact test with cells <5) and Wilcoxon rank-sum test for medians

Results from multivariable conditional logistic regression models relating NHL risk to TNFI use are reported in Table 3. After controlling for age, gender, and indication in the partially-adjusted model, risk of NHL was increased with TNFI ever-use (OR=2.09, 95% CI 1.29-3.40). Risk remained elevated with TNFI ever-use in the fully-adjusted model (Model 2), which additionally controlled for CCI score and use of oral corticosteroids, NSAIDs and csDMARDs during follow-up (OR=1.93, 95% CI 1.16-3.20). We observed a suggestive trend regarding duration of TNFI use and risk of NHL (P=0.05). Compared to never users, 1.5-2.5 years of TNFI use was associated with higher odds of NHL (OR=3.29, 95% CI 1.33-8.16) in the fully-adjusted model, while NHL risk was not elevated among the most persistent users of >2.5 years (OR=1.02, 95% CI 0.34-3.04). In addition, compared to never-users of DMARDs, users of TNFIs only had higher odds of developing NHL (OR 3.52, 95% CI 1.33-9.37), while use of only csDMARDs or both TNFIs and csDMARDs were not associated with elevated risk in fully-adjusted models. In subgroup analyses restricted to rheumatoid arthritis patients only (see Supplemental Table S1), our findings were similar in direction and statistical significance to our main approach (ever-use of TNFI in RA patients only, OR 1.90, 95% CI 1.09-3.32).

Table 3.

Conditional multivariable logistic regression models relating ever use of TNF inhibitors, duration of use and combination treatment with conventional DMARDs to risk of non-Hodgkin lymphoma

			Model 1a			Model 2b
	No. cases (%)	No. controls (%)	OR	95% CI	P	OR	95% CI	P
TNF inhibitor ever/never use
Never used TNF inhibitors	68 (67.3)	785 (79.8)	1.00	Ref.		1.00	Ref.
Ever used TNF inhibitors	33 (32.7)	199 (20.2)	2.09	(1.29, 3.40)	<0.01	1.93	(1.16, 3.20)	0.01
Duration of TNF inhibitor use
Never used TNF inhibitors	68 (67.3)	785 (79.8)	1.00	Ref.		1.00	Ref.
<1.5 years	12 (11.9)	70 (7.1)	1.91	(0.93, 3.91)	0.08	1.79	(0.89, 3.62)	0.10
1.5 to 2.5 years	13 (12.9)	61 (6.2)	3.33	(1.44, 7.71)	<0.01	3.29	(1.33, 8.16)	0.01
>2.5 years	8 (7.9)	68 (6.9)	1.25	(0.45, 3.44)	0.66	1.02	(0.34, 3.04)	0.97
			P-trend: 0.02			P-trend: 0.05
History of treatment with conventional DMARDs
Never used TNF inhibitors or conventional DMARDS	19 (18.8)	238 (24.2)	1.00	Ref.		1.00	Ref.
Conventional DMARDs only	49 (48.5)	547 (55.6)	1.14	(0.64, 2.03)	0.66	1.11	(0.61, 2.00)	0.74
TNF inhibitors only	11 (10.9)	43 (4.4)	3.71	(1.46, 9.44)	0.01	3.52	(1.33, 9.37)	0.01
Both conventional DMARDs and TNF inhibitors	22 (21.8)	156 (15.9)	1.97	(1.00, 3.90)	0.05	1.74	(0.86, 3.53)	0.12

^aModel 1 adjusted for age, gender and rheumatologic indication

^bModel 2 adjusted for age, gender, rheumatologic indication, Charlson comorbidity score, use of oral corticosteroids, use of prescription NSAIDs, use of conventional DMARDs

In Table 4, we report estimates on the risk of NHL associated with type of TNFI used. Use of etanercept was associated with increased risk of NHL (OR 2.73, 95% CI 1.40-5.33), while use of anti-TNF monoclonal antibodies was suggestive of higher risk of NHL (OR 1.77, 95% CI 0.87-3.58) but not statistically significant. In a secondary analysis comparing use of anti-TNF monoclonal antibodies to etanercept (reference), no difference in risk was observed between these types of TNFI (OR 1.10, 95% CI 0.42-2.85) (see Supplemental Table S2).

Table 4.

Conditional multivariable logistic regression models relating type of TNF inhibitor use to risk of non-Hodgkin lymphoma

			Model 1a			Model 2b
	No. cases (%)	No. controls (%)	OR	95% CI	P	OR	95% CI	P
TNF fusion protein (etanercept)
Never used any TNFI	68 (67.3)	785 (79.8)	1.00	Ref.		1.00	Ref.
Used TNF fusion protein (etanercept) only	13 (12.9)	67 (6.8)	2.71	(1.35, 5.45)	0.01	2.73	(1.40, 5.33)	<0.01
Anti-TNF monoclonal antibodies
Never used any TNFI	68 (67.3)	785 (79.8)	1.00	Ref.		1.00	Ref.
Used anti-TNF monoclonal antibodies only	17 (16.8)	95 (9.7)	2.17	(1.13, 4.16)	0.02	1.77	(0.87, 3.58)	0.11

^aModel 1 adjusted for age, gender and rheumatologic indication

^bModel 2 adjusted for age, gender, rheumatologic indication, Charlson comorbidity score, use of oral corticosteroids, use of prescription NSAIDs, use of conventional DMARDs

In sensitivity analyses (Table 5), we determined the impact of unmeasured confounding by disease severity on risk estimates in our study. To entirely account for our findings of higher risk with TNFI use, the prevalence of high rheumatologic disease severity indicating “step-up” therapy (DAS scores >5.1) would have to be three times greater in TNFI users, and high disease severity would have to be associated with a more than four-fold increased risk of NHL.

Table 5.

Deterministic sensitivity analysis of the TNF inhibitor and NHL odds ratio simulating differential levels of high rheumatologic disease severity^a by exposure status and the relative risk of NHL with high rheumatologic disease severity

Prevalence of Unmeasured Confounder: High Rheumatologic Disease Severity		Relative Risk for Unmeasured Confounder-Disease Association: High Rheumatologic Disease Severity and Risk of NHL
TNFI Exposed	TNFI Unexposed	RRcd
Pc1	Pc0	RR=1.0	RR=1.5	RR=2.0	RR=2.5	RR=3.0	RR=3.5	RR=4.0
		Odds ratio (OR) for TNF inhibitors and NHL risk with external adjustment for differential disease severity by exposure status
0.50	0.50	1.93	1.93	1.93	1.93	1.93	1.93	1.93
0.55	0.45	1.93	1.85	1.80	1.76	1.74	1.72	1.70
0.60	0.40	1.93	1.77	1.68	1.62	1.57	1.54	1.51
0.65	0.35	1.93	1.70	1.57	1.48	1.42	1.37	1.34
0.70	0.30	1.93	1.64	1.47	1.36	1.28	1.22	1.18
0.75	0.25	1.93	1.57	1.37	1.24	1.15	1.09	1.03

^aSimulates external adjustment for a hypothesized binary confounder-disease association of high disease activity score (DAS ≥5.1) in relation to risk of non-Hodgkin lymphoma

DISCUSSION

In summary, we observed an overall higher risk of NHL associated with ever-use of TNFIs in an insured, adult U.S. population with RA, PsA or AS. Findings were suggestive of a trend of increasing NHL risk with duration of TNFI use but failed to establish a clear dose-response effect. By TNFI type, TNF fusion protein (etanercept) was associated with greater risk of NHL, while increased NHL risk observed with anti-TNF monoclonal antibodies was not statistically significant. Our deterministic bias analysis suggested that our findings were unlikely due to unmeasured confounding by disease severity, or channeling bias, indicating a need for continued surveillance and awareness of the FDA black box warning and additional research on TNFIs and their biosimilar counterparts.

Our findings are not entirely consistent with results from a recent BSRBR-RA study¹⁸ and meta-analysis by Mariette, et al.,²⁰ both of which concluded that there was no elevated risk of lymphomas with TNFI use. Importantly, while these two reports evaluated the composite outcome of both Hodgkin and non-Hodgkin lymphoma, our study evaluated specifically only NHL, which is the focus of the FDA black box warning.²⁷ Furthermore, earlier use of TNFIs in patients with less severe rheumatologic disease is notably more prevalent in the U.S. compared to the U.K.²³ and less than 10% of RA patients in the UK receive biologics due to restrictive coverage.²⁴ Expert guidance and clinical consensus from the National Institute for Health and Clinical Excellence in the UK determines TNFI therapy to be indicated for severe, active rheumatoid arthritis by multiple DAS scores of 5.1 or greater.³⁵ Other recent reports outline the clinical importance of earlier treat-to-target approaches with TNFIs and biologic DMARDs to prevent disability and modify disease progression.^{40, 41} Inherent differences in these patient populations²⁴ may partly explain the inconsistent findings of TNFI use and NHL risk, making additional studies and follow up in recent time periods particularly important.

The meta-analysis by Mariette, et al. included five registry studies that evaluated lymphoma risk with TNFI use, including two U.S. registries [National Data Bank for Rheumatic Diseases (NDBRD) and Consortium of Rheumatology Researchers of North America (CORRONA)].²⁰ The pooled estimate of lymphomas from these studies was a relative risk of 1.11 (95% CI 0.70-1.51) and a standardized incidence ratio (SIR) of 2.55 (95% CI 1.93-2.17) compared to the general population. Among patients enrolled in the NDBRD from 1998-2005, odds ratios for lymphoma were 1.0 (95% CI 0.6-1.8) comparing TNFI use versus no use and 1.1 (95% CI 0.6-2.0) comparing TNFI + methotrexate versus methotrexate alone.¹⁵ Analysis of 19,608 RA patients enrolled in the CORRONA registry through 2013 found that TNFI use did not increase the incidence of lymphoma over an average of 3.1 years⁸ and disease severity was not related to lymphoma risk, a conclusion that is contrary to prior knowldege.⁶ Further inconsistency documented in these data are related to the significant differences between Hodgkin lymphoma and NHL, as well as within NHL subtypes. For example, subgroup analyses of the BSRBR-RA study showed suggestive increases in the risk for diffuse large B-cell lymphoma (hazard ratio [HR]=1.54, 95% CI 0.60-3.95) but somewhat lowered for Hodgkin lymphoma (HR=0.54, 95% CI 0.12-2.50).¹⁸ These patient registry studies, while prospective and real-world in nature, have some limitations including patient selection and retention in the registry, as well as under-ascertainment of outcomes when patients exit due to illness or more severe disease.⁴²

In our study, risk of NHL was 3.3 times higher among TNFI users who remained on therapy for 1.5-2.5 years, but not with longer-term use. These findings are in general consistent with those reported in the French RATIO registry study that TNFI duration of less than 2 years was independently associated with a higher risk of lymphoma in the French RATIO registry (OR=3.30, 95% 1.17-9.30).⁷ However, the Swedish registry study found no evidence of any duration-response relationship between TNFI use and overall malignancy.^{20, 43} There are several plausible explanations for this apparent non-linear trend with duration of use. The observation of no increased risk among more persistent users could be due to a depletion of susceptibles⁴⁴, where patients at highest risk discontinue TNFIs earlier due to nonspecific symptoms (e.g. fatigue, weight loss) in early disease. Further, healthy user bias⁴⁵, where patients who are more persistent on therapy also exhibit other healthy behaviors, could compose a stratum of the cohort at lower risk for NHL.

We found that the type of TNFI appeared to also influence the risk of NHL, where etanercept use was associated higher risk, while the greater observed risk with monoclonal antibody TNFI use was not statistically significant, although direct comparisons between TNFI types were limited. In the French RATIO registry study, risk of lymphomas was not elevated with etanercept (SIR 0.9, 95% CI 0.4-1.8), but increased with monoclonal antibody TNFI (SIR 3.7, 95% CI 2.6-5.3).¹⁶ In the BSRBR-RA study, there was no evidence of heterogeneity of effect by TNFI type.¹⁸ Although these studies, including ours, have limited power for subgroup analysis, it remains possible that individual anti-TNF monoclonal antibody agents might differentially impact NHL risk as dissimilar risk of cancers has been observed in other studies.⁴³ Unfortunately, the underlying drug mechanisms for excess NHL risk is unclear. With several biosimilars currently FDA-approved for the treatment of rheumatologic conditions and more in the foreseeable future^{46, 47}, continued surveillance of the safety of individual biosimilars is necessary. Unlike small molecule generics, the manufacturing process and the complexity of protein structures could result in differences in outcomes, even for the same biologic drug.⁴⁶

Determining a possibly causal link between NHL and TNFI use is complex, due in part to the rarity of the outcome, possible confounding from underlying disease severity and risk with other immunotherapies (including csDMARDs). We observed the highest risk in patients with a history of TNFI monotherapy use only, followed by users with exposure to both TNFI and csDMARDs, where odds of NHL were 3.5 and 1.7 times higher, respectively, compared to DMARD naive patients. It is unclear whether combination DMARD therapy may have led to further reductions in rheumatologic disease severity and therefore less inflammation, resulting in lower NHL risk. In a comparative safety study, overall cancer risk was elevated for methotrexate users compared with TNFIs, and specifically for lymphoma, there was a non-significant trend towards lower risk with methotrexate use (HR 0.15, 95% CI 0.01-2.19).¹³ Another possible explanation is that patients newly starting on DMARDs represent a high risk group, which raises important questions as early use of TNFI monotherapy has been increasing in the U.S.²⁵ Of note, the treat-to-target approach now recommended by international consensus^{40, 41, 48} may urge even more patients to start anti-TNF therapy earlier.

There is a lack of full understanding of the biologic actions of TNFIs and the pathway between TNFI-related immunosuppression and NHL. Suppression of disease activity in rheumatologic diseases could potentially reduce cancer risk by reducing uncontrolled inflammation.^{2, 3} Other recent research has shown that use of TNFIs in RA patients is associated with reduced activation and function of natural killer cells, which are a prominent line of defense against certain cancers⁴⁹ and further establishes a biologic rationale basis for the increased risk of NHL with TNFI use. Continued surveillance of lymphoma risk with TNFI use is needed, especially as case reports of lymphomas with TNFI use continue to surface.^50–52

Finally, some but not all studies with measures of clinical disease severity such as DAS scores explain the observed increased risk of NHL as a function of greater cumulative inflammatory activity, not treatment. In sensitivity analyses, we estimated that a minimum of three times higher prevalence of a potential confounder, DAS scores ≥5.1 indicating high rheumatologic disease severity, with four-fold greater risk of NHL would be needed to fully account for our results. This scenario is unlikely given that disease severity was not a predictor of higher lymphoma risk⁸ in the CORRONA Registry study and evidence indicates increasing TNFI use for less severe rheumatologic disease in the U.S.²³

Strengths and Limitations

This study has several strengths including sampling from a large cohort of patients in a U.S. health insurance-based database during a contemporary treatment period (2009 to 2015) with detailed information on medication use from pharmacy dispensing records and parenteral infusions, including duration of TNFI use, different TNFI types and concurrent therapies. There were also limitations. First, as with any study utilizing an administrative claims database, these data were not originally collected for the purposes of research and confounding by unmeasured factors is always possible. Specifically, we did not have information on the total duration or severity of rheumatologic disease. We attempted to account for this in our study design by selecting an active treatment cohort, requiring that patients were seen by a rheumatology provider for their index diagnosis to cohort entry and were receiving 2+ prescription medications or infusions directed at their underlying rheumatologic condition (versus having only ICD-9 diagnosis codes with no active treatment and seen only by primary care providers). This was also done, in part, to make the composition of past and current treatment in the underlying cohort similar to registry studies where patients were receiving active treatment.¹⁸ Further, we measured and adjusted for baseline and current treatments with NSAIDs, corticosteroids and csDMARDs as possible indications of greater duration and severity of rheumatologic disease. Also, while the algorithm used for case identification was validated for NHL, information on specific subtype could not be ascertained. Second, an alternative study design to our approach to investigate the safety of TNFIs is the active comparator, new user cohort study. In order to do this, we would need to compare treatment groups that consisted of patients newly initiating csDMARDs with and without TNFI therapy and also TNFI monotherapy. However, given the changing prescribing patterns in initiating these therapies and the rare occurrence of non-Hodgkin lymphoma, this would leave our study relatively lacking in power to detect a true association if one existed. Third, while pharmacy dispensing and medication administration data are a reproducible and reliable way to determine medication use in this insured population, patients may have filled the prescription but not taken it as directed. However, in prior analyses of TNFI use in this population, we determined that while the adherence to TNFIs varies, persistence to self-administered drugs in this class is relatively consistent.⁵³ Thus, while our findings are supportive of an association with increased NHL risk, this study also raises multiple questions, particularly with respect to an interaction with rheumatologic disease severity and patient subgroups at risk (e.g., biologic DMARD initiators and long-term users).

Conclusion

We observed a positive association between TNFIs and risk of NHL. These biologic DMARD therapies result in substantial improvement in patients with rheumatologic conditions such that the benefits of treatment likely outweigh the risk of this rare but serious adverse outcome.^54–57 Despite this, the continued surveillance and awareness of the FDA black box warning⁴ for lymphoma risk are warranted as trends in the earlier and long-term use of TNFIs continue to increase, new TNFIs with different pharmacologic characteristics become available and with TNFI biosimilar products forthcoming.

Novelty

Based on limited evidence, the U.S. Food and Drug Administration issued a black box warning for the use of tumor necrosis factor-alpha inhibitors and risk of non-Hodgkin lymphoma in 2009 and alerted the public regarding case reports of rare non-Hodgkin lymphoma subtypes in 2011. In a large U.S. health insurance-based cohort of adults with rheumatologic conditions between 2009 and 2015, tumor necrosis factor-alpha inhibitors were associated with increased non-Hodgkin lymphoma risk. Our findings also suggested increased risk with longer duration of use and differences by type of agent.

Impact

Our study supports the current black box warning for risk of non-Hodgkin lymphoma with use of tumor necrosis factor-alpha inhibitors. Continued awareness of this rare but serious adverse outcome is warranted with new agents and biosimilar products in this drug class forthcoming.