The effect of methotrexate and anti–tumor necrosis factor therapy on the risk of lymphoma in rheumatoid arthritis in 19,562 patients during 89,710 PERSON-YEARS of observation


  • The National Data Bank for Rheumatic Diseases has conducted safety registries for Centocor, Sanofi-Aventis, and Bristol-Meyers Squibb



To ascertain the relationship between anti–tumor necrosis factor (anti-TNF) therapy, methotrexate (MTX), and the risk of lymphoma in patients with rheumatoid arthritis (RA). This report updates our previous report during 29,314 person-years of followup.


Participants in the National Data Bank for Rheumatic Diseases (NDB) longitudinal study of long-term outcomes of RA completed semiannual questionnaires from 1998 through 2005, during 89,710 person-years of followup. Lymphoma reports were validated by medical records. The association between lymphoma and treatment was investigated using conditional logistic regression, adjusted for severity and demographic covariates.


Of the 19,591 participants, 55.3% received biologic agents and 68.0% received MTX while enrolled in the NDB. The lymphoma incidence rate was 105.9 (95% confidence interval [95% CI] 86.6–129.5) per 100,000 person-years of exposure. Compared with the SEER (Surveillance, Epidemiology, and End-Results) lymphoma database, the standardized incidence ratio was 1.8 (95% CI 1.5–2.2). The odds ratio (OR) for lymphoma in patients who received anti-TNF therapy compared with patients who did not receive anti-TNF therapy was 1.0 (95% CI 0.6–1.8 [P = 0.875]). The OR for lymphoma in patients who received anti-TNF plus MTX therapy compared with patients who received MTX treatment alone was 1.1 (95% CI 0.6–2.0 [P = 0.710]). Infliximab and etanercept considered individually also were not associated with a risk of lymphoma.


In a study of lymphoma in 19,591 RA patients over 89,710 person-years of followup, which included exposure to anti-TNF therapy in 10,815 patients, we did not observe evidence for an increase in the incidence of lymphoma among patients who received anti-TNF therapy.

In 2004, we published data on malignant lymphoma in rheumatoid arthritis (RA), based on 29 lymphomas that occurred during 29,314 person-years of followup (1). The purpose of the current article is to update that report by extending it to 89,710 person-years, during which 95 lymphomas were identified, in order to address the issue of the association between lymphoma risk and treatment with an anti–tumor necrosis factor (anti-TNF) agent and methotrexate (MTX). In the original report, we concluded that lymphomas were increased in RA (standardized incidence ratio [SIR] 1.9, 95% confidence interval [95% CI] 1.3–2.7), and that the SIR was greatest for anti-TNF therapies. However, the risk difference between therapies was slight, and CIs for the treatment groups overlapped.

Since the publication of our original article, additional reports from registries, longitudinal studies, and meta-analyses have become available. Those studies and others related to RA treatment and lymphoma risk were recently reviewed in detail (2). Observational studies germane to the relationship between lymphoma and anti-TNF therapy are described in Table 1. Using several large observational cohorts linked to the Swedish Cancer Register (3), Askling et al noted the overall SIR for lymphoma to be 1.9–2.0 in 2 cohorts that did not receive anti-TNF therapy. Compared with no anti-TNF treatment, the adjusted relative risk of lymphoma associated with anti-TNF therapy was 1.1 (95% CI 0.6–2.1).

Table 1. Lymphoma incidence rates, SIRs, and RRs associated with anti-TNF therapy*
Author (ref.)Study yearsStudy typeNo. of lymphomasNo. of patientsPerson-years of followupIncidence per 100,000 patient-years (95% CI)SIR (95% CI)RR (95% CI)
  • *

    SIR = standardized incidence ratio; RR = relative risk; 95% CI = 95% confidence interval; RA = rheumatoid arthritis.

  • Versus reference group.

  • Versus non–anti–tumor necrosis factor (anti-TNF)–treated patients.

Wolfe and Michaud (1)1998–2002Observational2918,57429,31498.9 (69–142)1.9 (1.3–2.7)
Askling et al (3)1990–2003Observational (inpatient registry)31953,067297,1021.9 (1.7–2.1)Reference
Askling et al (3)1990–2003Observational (early RA cohort)113,7033,7032.0 (1.0–3.5)0.8 (0.4–1.4)
Askling et al (3)1990–2003Observational (anti-TNF cohort)94,1609,7152.9 (1.3–5.5)1.1 (0.6–2.1)
Geborek et al (4)1999–2002Observational71,5575,58711.5 (3.7–26.9)4.9 (0.9–26.2)
Franklin et al (6)1990–2004Observational91,2379,27497.0 (44.4–184.2)2.9 (1.3–5.6) 
Wolfe and Michaud (current study)1998–2006Observational9519,59189,710105.9 (86.6–129.5)1.8 (1.5–2.2)1.0 (0.6–1.8)

In a partially cohort-based case–control study of 1,557 patients, Geborek et al noted an SIR of 11.5 and a relative risk of 4.9 (95% CI 0.9–26.2) for anti-TNF–treated patients compared with non–anti-TNF–treated patients (4). Franklin et al, in an accompanying editorial, considered methodologic problems with this report, including the possibility of confounding by indication, latency, and a low rate of lymphoma in the control population (5).

Franklin et al reported data from the Norfolk Arthritis Register (NOAR) in the UK, in which the use of anti-TNF therapy was rare (6). The SIR for lymphoma was 2.9 (95% CI 1.3–5.6), and the risk of lymphoma was associated with RA disease activity.

Baecklund et al reported on 378 consecutive patients from a population-based cohort of 74,651 Swedish patients with RA in whom lymphoma developed between 1964 and 1995 and compared them with 378 control patients (7). Those investigators observed that standard nonbiologic treatments did not increase the risk of lymphoma, and that high inflammatory activity rather than treatment was the major risk determinant.

In contrast, a recent meta-analysis of infliximab and adalimumab clinical trials by Bongartz et al identified 10 lymphomas in 3,493 anti-TNF–treated patients compared with no lymphomas in 1,512 non–anti-TNF–treated patients (8). That report has been both criticized (9) and defended (10).

In the current analysis, we used conditional logistic regression to analyze treatment risks in 19,562 patients, including 43 lymphomas occurring in 10,815 anti-TNF–treated patients and 50 lymphomas in 8,747 non–anti-TNF–treated patients.



Patients in the study were participants in the National Data Bank for Rheumatic Diseases (NDB) longitudinal study of the outcomes of RA, who completed semiannual questionnaires in the period from 1998 through 2005. Patients were recruited on an ongoing basis from the practices of US rheumatologists and are followed up prospectively by NDB staff, using semiannual, detailed, 28-page questionnaires, as previously described (1). The diagnosis of RA was made by the patient's rheumatologist. To be eligible for this study, patients were required to have completed at least 2 semiannual questionnaires and not to have been diagnosed as having lymphoma prior to study entry.

At the time of enrollment in the NDB, we determined specific treatments taken by participants prior to enrollment as well as previous medical history, including malignancies. At the time of each semiannual questionnaire assessment, we recorded demographic, disease severity, treatment, and malignancy variables.

Disease severity variables used as baseline variables in this report included the number of disease-modifying antirheumatic drugs (DMARDs) taken prior to enrollment, use of prednisone at enrollment, and Health Assessment Questionnaire (HAQ) disability scores (11). Sociodemographic variables included age at enrollment, lifetime history of smoking, education level, and health insurance characteristics.

Patients use the semiannual questionnaires to report use of all medications in the previous 6 months. We defined treatment with a DMARD to be treatment with any one of the following drugs: leflunomide, auranofin, azathioprine, sulfasalazine, cyclosporine, cyclophosphamide, injectable gold, minocycline, penicillamine, hydroxychloroquine, and MTX.

All reports of malignancy, including reports of lymphoma, were followed up by contacting the patient for specific details, and then were followed up further by a request for hospital records or physician confirmation. We also searched the National Death Index (NDI) (12) annually and received reports of deaths from family and friends of the participants. Twenty lymphomas were identified from death records. Of the 95 lymphomas identified, 90 were supported by medical evidence. The other 5 were judged likely to be lymphomas, but additional confirmatory medical evidence was not available to confirm or refute the diagnosis. This situation occurred when consent for additional medical information could not be obtained or when requests for medical records were refused.

Entry criteria were met by 19,591 patients, and their data were used to determine the lymphoma SIR and the lymphoma incidence rate. Twenty-nine patients had incomplete treatment data. Therefore, analyses of treatment associations were confined to the 19,562 patients for whom complete data were available. To determine expected rates of lymphoma, we used the US SEER (Surveillance, Epidemiology, and End-Results) database as a comparison population (13). The SEER program of the National Cancer Institute is an authoritative source of information on cancer incidence and survival in the US. SEER currently collects and publishes cancer incidence and survival data from population-based cancer registries covering ∼26% of the US population. We used age and sex categories from the SEER database for Hodgkin's lymphoma (International Classification of Diseases, Ninth Revision [ICD-9] 201) and non-Hodgkin's lymphoma (ICD-9 200, 202.0–202.2, 202.8–202.9) to estimate the SIR for lymphoma in the RA study population compared with the US population.

Statistical analysis.

We calculated lymphoma incidence rates utilizing Poisson CIs. Exposure time (i.e., time in the study) was calculated as the time from entry into the cohort through the last observation period for patients without lymphoma and to the time of lymphoma diagnosis for patients with lymphoma. In instances in which lymphoma was diagnosed at the time of death, exposure time was taken from the time of last observation in the NDB. In sensitivity analyses, exposure time extending to the time of death was also examined, and analyses were also carried out with exclusion or lymphomas identified by death records only. We included delayed death record case identification because case followup in the early years of the data bank was not as effective as it is currently.

Use of specific treatments or groups of treatments (e.g., anti-TNF agents) was defined as receiving that treatment or group of treatments while participating in the NDB. Treatments that were administered after the development of lymphoma were not counted.

Comparisons between treatments were performed using conditional logistic regression. By conditioning on the year of study entry and the last year of study participation, patients were matched for both calendar time of entry and length of time in the study. For additional details of the methodology of the conditional logistic regression analyses and sensitivity analyses performed, see the Discussion (below).

Data were analyzed using Stata version 9.2 software (Stata Corporation, College Station, TX). Statistical significance was set at the 0.05 level, CIs were established at 95%, and all tests were 2-tailed. P values were not adjusted for multiple comparisons.


Of the 19,591 patients enrolled, 55.3% received biologic agents while participating in the NDB, and 68.0% received MTX during this period (Table 2). At study initiation (baseline), 57.7% of patients were receiving MTX, and 45.7% were receiving prednisone. The mean ± SD lifetime count of DMARDs at study initiation was 2.3 ± 1.6.

Table 2. Demographic and clinical characteristics of the 19,591 participants*
  • *

    Except where indicated otherwise, values are the percentage. NDB = National Data Base; DMARD = disease-modifying antirheumatic drug; HAQ = Health Assessment Questionnaire.

Age at baseline, mean ± SD years59.0 ± 13.2
Male sex22.8
Education, mean ± SD years13.2 ± 2.2
Disease duration at study start, mean ± SD years14.1 ± 12.1
Ever smoker55.4
Biologic agent use ever in NDB55.3
Infliximab use ever in NDB40.3
Etanercept use ever in NDB19.2
Adalimumab use ever in NDB7.6
Methotrexate use ever in NDB68.0
Baseline lifetime DMARD count (0–11 scale)2.3 ± 1.6
Prednisone therapy at baseline45.7
Methotrexate therapy at baseline57.7
Baseline HAQ score, mean ± SD (0–3 scale)1.1 ± 0.7

NDB semiannual assessments occurred during 89,710 person-years of followup. The lymphoma incidence rate was 105.9 (95% CI 86.6–129.5) per 100,000 person-years of exposure. Compared with the SEER lymphoma database, the SIR was 1.8 (95% CI 1.5–2.2) based on 52.2 expected cases and 95 actual cases. We also calculated the SIR after excluding lymphomas identified with delays greater than 6 months after followup capture time. This resulted in a change in the SIR to 1.7 (95% CI 1.4–2.1).

Considering all patients treated with anti-TNF agents during NDB followup (mean 3.7 years, median 3.4 years), regardless of concomitant therapy, there was no increase in the risk of lymphoma (odds ratio [OR] 1.0, 95% CI 0.6–1.8 [P = 0.875]) (Table 3). Removing covariate control for inflammatory activity (HAQ score, number of prior DMARDs, and prednisone use) led to an increase in the OR to 1.3 (95% CI 0.8–2.1 [P = 0.333]), indicating that covariate control was effective. In addition, restricting the sample to patients who had never received anti-TNF therapy prior to participating in the NDB did not increase the risk associated with anti-TNF therapy (OR 0.9, 95% CI 0.5–1.5 [P = 0.784]). We also investigated whether the risk of having lymphoma was greater among TNF-treated patients during the first year of treatment. The OR for first-year lymphomas compared with later lymphomas was 1.0 (95% CI 0.4–3.1 [P = 0.937]).

Table 3. Association of RA treatment with lymphoma*
ModelTreatmentComparisonOR (95% CI)P
  • *

    For models 1 and 2, n = 10,364. For model 3, n = 7,916; patients treated with etanercept or adalimumab were excluded from analysis. For model 4, n = 6,259; patients treated with infliximab or adalimumab were excluded from analysis. For model 5, n = 5,110; patients treated with infliximab or etanercept were excluded from analysis. RA = rheumatoid arthritis; OR = odds ratio; 95% CI = 95% confidence interval; anti-TNF = anti–tumor necrosis factor; MTX = methotrexate.

  • Determined by conditional (fixed-effects) logistic regression and adjusted for initial age, sex, education level, lifetime history of smoking, health insurance, and initial values for age, Health Assessment Questionnaire score, prednisone use, and lifetime count of disease-modifying antirheumatic drugs. See Materials and Methods for details.

1Anti-TNFNo anti-TNF1.0 (0.6–1.8)0.875
2MTX and anti-TNFAll other treatments1.4 (0.7–2.9)0.331
2Anti-TNF only, no MTXAll other treatments0.8 (0.4–2.0)0.711
2MTX only, no anti-TNFAll other treatments1.3 (0.7–2.4)0.465
2MTX and biologicOnly MTX1.1 (0.6–2.0)0.710
3MTX and infliximabAll other treatments1.5 (0.6–3.4)0.337
3Infliximab only, no MTXAll other treatments1.0 (0.4–2.9)0.975
3MTX only, no infliximabAll other treatments1.3 (0.7–2.5)0.440
3MTX and infliximabOnly MTX1.2 (0.6–2.3)0.684
3All infliximabAll other treatments1.2 (0.6–2.2)0.646
4MTX and etanerceptAll other treatments1.1 (0.4–3.1)0.862
4Etanercept only, no MTXAll other treatments0.5 (0.1–2.2)0.456
4MTX only, no etanerceptAll other treatments1.3 (0.7–2.6)0.429
4MTX and etanerceptOnly MTX0.8 (0.3–2.1)0.710
 All etanerceptAll other treatments0.7 (0.3–1.6)0.422
5MTX and adalimumabAll other treatments5.6 (1.1–29.0)0.041
5Adalimumab only, no MTXAll other treatments0.00.998
5MTX only, no adalimumabAll other treatments1.3 (0.7–2.5)0.455
5MTX and adalimumabOnly MTX4.3 (0.9–21.2)0.072
 All adalimumabAll other treatments4.5 (0.9–23.1)0.064

To better evaluate the effect of anti-TNF therapy, we categorized participants into 5 mutually exclusive groups, as follows: simultaneous use of anti-TNF therapy and MTX, nonsimultaneous use of anti-TNF therapy and MTX, use of MTX and never use of anti-TNF therapy, use of anti-TNF therapy and never use of MTX, and use of treatments that did not include anti-TNF therapy or MTX. Nonsimultaneous use of anti-TNF therapy and MTX was rare. For ease of interpretation, patients who received anti-TNF therapy and MTX nonsimultaneously were combined with patients who received anti-TNF therapy only (Table 3). In addition, to study the effect of individual anti-TNF therapies, we removed patients from analysis who ever received an additional anti-TNF therapy different from the one under study in the second, third, and fourth models shown in Table 3.

Model 2 demonstrated no association of anti-TNF therapy and MTX with lymphoma. Compared with MTX alone, the combination of anti-TNF therapy and MTX did not increase the risk of lymphoma (OR 1.1, 95% CI 0.6–2.0 [P = 0.710]). In addition, after omitting from the analyses patients who had received anti-TNF therapy, the OR for lymphoma associated with MTX compared with other DMARDs (n = 3,083) was 1.3 (95% CI 0.6–2.7 [P = 0.538]).

Models 3 and 4 for infliximab and etanercept, respectively, showed no association of these treatments with lymphoma. The OR for the association of any infliximab treatment with lymphoma was 1.2 (95% CI 0.6–2.2 [P = 0.646]), and the OR for the association of any etanercept treatment was 0.7 (95% CI 0.3–1.6 [P = 0.422]). Model 5 showed a statistically significant association between adalimumab treatment and lymphoma. However, this association was based on 56 cases and only 2 lymphomas in the adalimumab-treated group. The limited number of cases reflects the study requirement that patients receiving specific anti-TNF therapies should not have been treated with other anti-TNF therapies first. When this requirement was relaxed for adalimumab, the number of cases expanded to 1,482, and the OR for adalimumab versus all other therapies was 1.2 (95% CI 0.3–5.10 [P = 0.812]).


The major threat to the validity of observational treatment studies is nonrandom assignment to treatment (channeling bias, confounding by indication). Specifically, if RA severity is related to the risk of developing lymphoma (2, 6, 7) and if patients with the most severe RA are more likely to receive anti-TNF therapy, then the observed lymphoma risk could represent any risk associated with treatment plus the risk associated with more severe RA. In addition to this potential bias, RA severity and the consequent probability of anti-TNF treatment vary over time. For example, persons who received anti-TNF therapy immediately after this treatment became available often had more severe RA than those who received it several years after it was marketed. Beyond severity issues and time-dependent biases, prescription of anti-TNF therapy could be related to age, access to treatment related to medical insurance, and other sociodemographic factors.

To control for these biases, we modeled lymphoma risk with conditional logistic regression. In our preliminary analyses, we assigned patients to 1 of 153 groups based on the 6-month calendar period in which they entered the data bank and the 6-month period in which they were last observed in the data bank (for example, entry in the first half of 2000 and last observation in the second half of 2004). Conditional logistic regression effectively compares patients only with others who are in the same time-based group. It controls, in part, for exposure time, differential severity over time, characteristics of participation and nonparticipation, and time-related risk of prescription of anti-TNF therapy. However, if patients in any group do not experience lymphoma, that group is excluded from conditional logistic regression analysis, because it is not possible to compare lymphoma cases with non-cases if everyone in a group represents a non-case. Because of this characteristic of conditional logistic regression, the more groups that are present, the smaller the ultimate sample size will be. In the analyses of this study, we first examined the data using fine control in 153 semiannual groups. We then also examined the data using 45 annual groups. The results were similar, but because the sample size after removal of groups without lymphoma cases was larger (n = 10,364 [9,197 observations dropped] versus n = 6,460 [13,101 observations dropped]), we report analyses based on the larger yearly groupings.

In addition to the adjustments inherent in conditional logistic regression, we adjusted for differences in baseline disease severity by including the number of previous DMARDs, initial prednisone use, initial HAQ score, and RA duration. Prednisone use at study initiation is a marker for severity, as is the lifetime count of DMARDs and the initial HAQ score. Sociodemographic factors that might influence prescriptions or lymphoma risk (age, sex, education level, and lifetime smoking history) were also included in the model.

As shown in Table 3, we did not find supporting evidence for a link between anti-TNF therapy or MTX and the risk of lymphoma. Considering patients treated with anti-TNF therapy versus those not receiving that therapy (model 1), the OR for lymphoma was 1.0 (95% CI 0.6–1.8). This result is similar to the relative risk of 1.1 (95% CI 0.6–2.1) noted by Askling et al using the Swedish register (3). When model 1 was run without covariates (except age and sex), the OR increased to 1.3 (95% CI 0.8–2.1 [P = 0.334]).

The more detailed analyses of model 2 allow the consideration of combinations of therapies as they are used in ordinary clinical practice. Of particular interest, the use of anti-TNF therapy and MTX compared with MTX alone showed an OR of 1.1 (95% CI 0.6–2.0). This shows that the most common way to administer anti-TNF therapy (anti-TNF plus MTX) does not increase the lymphoma risk compared with the most common DMARD therapy (MTX without anti-TNF therapy). Because these results might reflect specific anti-TNF therapies, we analyzed the data for infliximab and those for etanercept separately (models 3 and 4). Results were similar, and we still did not find evidence of an association between anti-TNF therapy and lymphoma.

A series of sensitivity analyses were conducted to test the robustness of the study results. None of the analyses led to clinically or statistically significant changes in results. In these sensitivity analyses, we separately 1) excluded patients who entered the NDB as part of the leflunomide safety registry, 2) added a dummy variable to account for leflunomide participation, 3) added 3 dummy variables to account for membership in safety registries, and 4) excluded patients whose lymphoma was identified only by analysis of death records.

The results of this study are in agreement (Table 1) with the linked Swedish registry studies (3) but are in disagreement with the observational data described by Geborek et al (4), who identified 5 lymphomas among 757 anti-TNF–treated RA patients (1,603 person-years) and 2 lymphomas among 800 patients in an RA comparison cohort (3,940 person-years). The SIR for the comparison group was 1.3. Franklin et al, in an accompanying editorial, considered methodologic problems with that report, including the possibility of confounding by indication, latency, and a low rate of lymphoma in the control population (5). Our data also differ from the results of the clinical trial meta-analysis by Bongartz et al, who identified 10 lymphomas in 3,493 anti-TNF–treated patients compared with no lymphomas in 1,512 non–anti-TNF–treated patients (8). Possible problems with these results were discussed in an accompanying editorial by Dixon and Silman (9).

One limitation of the current study is that a very small number of deaths that are unknown to the NDB but will be identified by future reports from the NDI have not yet been identified, because NDI reporting is typically delayed by 18 months. This number is likely to be very small (∼2), because we also learn about deaths from physicians and families. This bias would increase the overall SIR very slightly but should have no effect on the relationship between treatment and lymphoma risk.

It is also possible that we failed to identify some cases of lymphoma. This might occur because patients who stop participating in the NDB might do so because of concomitant illness. The NDB conducts exit interviews and contacts physicians for followup to prevent this bias. The SIR of 1.8 we obtained is similar to the SIR obtained in the large Swedish registry of 53,067 RA patients (3). Although Franklin et al found an SIR of 2.9 (95% CI 1.3–5.6) in NOAR (6), they also noted a lymphoma incidence of 97.0 (95% CI 44.4–184.2) per 100,000 patient-years. The incidence of lymphoma in the NDB was 109.3 (95% CI 90.1–132.6). If some cases were missed, however, this would not alter the treatment-related risk of lymphoma.

In this study, we controlled for disease activity using entry values for the HAQ, the number of prior DMARDs, and use of prednisone. When these variables were removed, the OR increased from 1.0 to 1.3, indicating effective covariate control. This was to be expected, because summary patient variables (14) are correlated with the Disease Activity Score at r ∼0.6 (15). We did not have access to information about acute-phase reactants or joint counts, and it is possible we did not have full covariate control. If this were the case, we would expect that the “true” risk of lymphoma associated with anti-TNF therapy would be even less than the risk we identified.

The results of this study provide strong evidence that lymphoma is not increased among RA patients treated with anti-TNF agents in routine clinical practice, and is in agreement with the linked cancer registry reports from Sweden by Askling et al (3). These results are substantially different from results of clinical trials (8). It has been suggested that one explanation for this discrepancy is that the effect seen in clinical trials might represent an “unmasking” of cases that will be identified later in observational studies.

In summary, in a study of lymphoma in 19,562 patients over 89,710 person-years of followup, which included anti-TNF exposure in 10,815 patients, we did not find evidence for an increase in lymphoma in anti-TNF–treated patients.


Dr. Wolfe had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Wolfe, Michaud.

Acquisition of data. Wolfe, Michaud.

Analysis and interpretation of data. Wolfe, Michaud.

Manuscript preparation. Wolfe.

Statistical analysis. Wolfe, Michaud.


The data for this study were collected and analyzed by the authors, who also wrote the manuscript without assistance from others. According to our safety registry contractual agreement with Centocor, the completed manuscript was reviewed by Centocor. No changes in the manuscript were made after their review. We declare we have no conflict of interest.