Tumor necrosis factor α antagonist use and cancer in patients with rheumatoid arthritis

Abstract

Objective

Concerns persist about a possible association between tumor necrosis factor α (TNFα) antagonist treatment and development of cancers in patients with rheumatoid arthritis (RA). This study was undertaken to estimate the association between treatment with biologic disease-modifying antirheumatic drugs (DMARDs) and development of cancer in patients with RA.

Methods

We conducted a cohort study pooling administrative databases from 2 US states and 1 Canadian province. A cohort of patients who had received a diagnosis of RA on ≥1 occasion and had been prescribed DMARDs was identified. We categorized patients with a prescription for a biologic DMARD as biologic DMARD users, and those with a prescription for methotrexate (MTX) but no biologic DMARD as MTX users. We used time-varying propensity scores to adjust for the large number of possible confounders and stratified proportional hazards regression to estimate the effects of biologic DMARDs on cancer. The primary end points were hematologic malignancies (lymphoma, multiple myeloma, and leukemia) and common solid tumors (colorectal, lung, stomach, breast, prostate, uterine, ovarian, urinary tract/bladder, and melanoma).

Results

The pooled cohort included 1,152 biologic DMARD users and 7,306 MTX users. We identified 11 hematologic malignancies and 46 solid tumors during 2,940 person-years of biologic DMARD use, and 88 hematologic malignancies and 558 solid tumors during 30,300 person-years of MTX use. Comparing biologic DMARD users with MTX users, the propensity score–adjusted pooled hazard ratio was 1.37 (95% confidence interval 0.71–2.65) for hematologic malignancies and 0.91 (95% confidence interval 0.65–1.26) for solid tumors.

Conclusion

Our results indicate that users of biologic agents are unlikely to have a substantial increase in the risk of hematologic malignancies and solid tumors as compared with MTX users. Despite the use of large combined data sets, studying the effect of an infrequent exposure (biologic DMARDs) on rare diseases (hematologic malignancies) remains a challenge.

The risk of cancer, especially lymphoproliferative malignancies such as non-Hodgkin's lymphoma and multiple myeloma, has consistently been reported as being higher in patients with rheumatoid arthritis (RA) than in the general population (1–10). Although some of the risk appears to be linked to the immunodysregulation of RA, the use of disease-modifying antirheumatic drugs (DMARDs) also may play a role (8, 11–14).

The tumor necrosis factor α (TNFα) antagonists, a type of biologic DMARD with profound immunoregulatory effects, have had a significant impact on the treatment of RA. However, safety concerns regarding these drugs arose after the Food and Drug Administration's postmarketing spontaneous adverse event reporting system (MedWatch) received reports of lymphoma. Brown et al reviewed MedWatch reports of 26 cases of lymphoproliferative disorders that occurred following treatment with either of 2 TNFα antagonists, etanercept or infliximab. The extrapolated crude incidence of lymphoproliferative disorders was 19.9 per 100,000 person-years for etanercept and 6.6 per 100,000 person-years for infliximab (15). Although these rates were smaller than that in the general population, features of the cases were of concern: in 54% of the patients who developed lymphoma, it was detected within 8 weeks after initiation of TNFα antagonist treatment, and in 2 of these patients, it remitted after cessation of TNFα antagonist treatment.

A recent study compared the incidence of lymphoma in a cohort of patients with RA and that in the general population (16). The study documented increased risk of lymphoma in RA patients compared with the general population, and the risk was greatest in patients treated with TNFα antagonists. The authors concluded that the increased risk in the TNFα antagonist–treated group might reflect the fact that patients with the highest risk of lymphoma receive TNFα antagonists. Thus, they were unable to determine whether the increase in standardized rate ratios was because of RA or a true risk associated with the drugs. Another study combined a national cohort of Swedish patients with a registry of users of anti-TNFα drugs and showed a 5-fold increase of lymphoma incidence, but not of overall cancer incidence, among RA patients exposed to TNFα antagonists compared with those never treated with these drugs (17). Several methodologic issues, including insufficient control for confounding by indication, surveillance bias, and failure to detect an increased incidence of lymphoma in the unexposed RA population may have resulted in this apparent increase in the risk (17, 18). Thus, it remains unclear whether biologic DMARDs are associated with an increased risk of lymphoma or other malignancies in RA patients.

In the present study, pooling 3 large cohorts, we estimated the increase in risk of hematologic malignancies and common solid tumors in RA patients who received biologic DMARDs as compared with those treated with methotrexate (MTX).

PATIENTS AND METHODS

Study patients and data sources.

We conducted a cohort study using health care utilization databases from the US and Canada. The study population consisted of subjects from 3 data sources: 1) Medicare beneficiaries enrolled in the Pharmaceutical Assistance Contract for the Elderly in Pennsylvania from January 1, 1994 through December 31, 2004, 2) Medicare beneficiaries enrolled in the Pharmaceutical Assistance to the Aged and Disabled program or Medicaid in New Jersey from January 1, 1994 through December 31, 2004, and 3) all residents of British Columbia, Canada who were 18 years or older, from January 1, 1996 through December 31, 2003. Both drug benefit programs in Pennsylvania and New Jersey provide comprehensive pharmacy coverage with a small copayment or no copayment. Canada has a national insurance system that provides comprehensive coverage for health care, including pharmacy benefits for the elderly and disabled. The dispensing of prescription drugs in a community pharmacy is recorded for the entire population independent of insurance coverage. These data sources provide basic demographic information, as well as coded diagnostic, procedural, and pharmacy dispensing information, with high accuracy (19–21). The Institutional Review Boards of the Brigham and Women's Hospital and University of Victoria approved this study, and data use agreements were established. All potentially traceable personal identifiers were removed from the data prior to analysis, to protect patients' privacy.

In the databases, we identified a cohort of subjects age ≥65 years in the US and Canada who had at least 1 claim with a diagnosis of RA and who were dispensed at least 1 prescription of any DMARD or corticosteroid after the first RA diagnosis during the study period. We excluded subjects who had a diagnosis of any cancer (except nonmelanoma skin cancer) or human immunodeficiency virus infection.

DMARD exposure and potential confounders.

All patients included in the study had been prescribed a biologic DMARD (etanercept, infliximab, adalimumab, or anakinra) or MTX. Cohort followup started at the time of the first prescription of a biologic DMARD or MTX during the study period. Some patients contributed person-time to multiple exposure categories. Patients who were receiving both a biologic DMARD and MTX at the same time were categorized as biologic DMARD users. We assumed that the effect of the biologic DMARDs persists from the initiation of its use until the end of followup, testing the hypothesis that the effect of these agents on cancer is both profound and long term (16). In a secondary analysis, we assumed a 180-day induction period and excluded that period from the followup time.

We considered potential confounders, measured using diagnosis and procedure codes and/or prescription information in the data sets. These included demographic variables, documented risk factors for cancers, factors associated with severity of RA, health care utilization, other major comorbid conditions, and empiric diagnosis/procedure codes that may have affected prescription of DMARDs and/or development or detection of cancers. A list of all covariates tested for, in each of these general categories, is available upon request from the corresponding author.

Study end points.

Subjects were censored at the time of 1) death, 2) loss of eligibility for health care benefits, 3) end of the study period, or 4) occurrence of stated end points. The end points included development of solid tumors such as cancers of the colorectum, lung, stomach, breast, prostate, uterus, ovary, urinary tract/bladder, and kidney, and melanoma, or hematologic malignancies such as lymphoma, multiple myeloma, and leukemia. Lymphoproliferative malignancies such as non-Hodgkin's lymphoma, chronic lymphocytic leukemia, and multiple myeloma are a subset of hematologic malignancies, which were also examined as a separate outcome in the study. Using coded information on diagnoses, procedures, and prescribed medications, we developed various claims-based definitions of the incident cancer. In a validation study linking claims data with population-based cancer registry data in Pennsylvania, we found that a cancer definition based on 2 repeated codes for a cancer, or diagnosis codes plus cancer-related procedure codes (see below) had a specificity of ≥99% and a sensitivity of 80% for most types of cancer. The event date for each case was defined as the first date a cancer diagnosis appeared in the claims data.

Health care utilization data–based definitions for incident cancers.

In the primary analyses, a subject was classified as having an incident cancer if any of the following conditions were met: 1) ≥1 diagnosis of cancer plus any diagnosis or procedure codes related to complications of cancer or palliative care within 2 weeks, followed by another diagnosis of cancer within 12 months; 2) at least 1 diagnostic procedure with biopsy followed by ≥2 diagnoses of cancer on at least 2 different occasions within 12 months (recorded on different dates from the procedures); 3) ≥1 diagnosis of cancer plus any cancer-related surgery during the same hospitalization and/or visit; 4) ≥1 diagnosis of cancer plus any cancer chemotherapy during the same hospitalization and/or visit; 5) ≥1 diagnosis of cancer plus any radiation therapy during the same hospitalization and/or visit; 6) ≥1 diagnosis of cancer plus hematopoietic cell transplantation during the same hospitalization and/or visit; 7) ≥1 diagnosis of cancer plus oral chemotherapy dispensed within 2 weeks after the diagnosis; 8) ≥2 diagnoses of cancer on at least 2 different occasions within 2 months.

Data analysis.

We used Cox proportional hazards regression to estimate the unadjusted and age/sex-adjusted effects of biologic DMARDs on cancer incidence. To control for confounding by indication due to RA severity, we reduced all potential confounders into a propensity score that varied with time (22). The details on our method for estimating the time-varying propensity score are shown in Appendix A. Hazard ratios (HRs) adjusted for all possible confounders were obtained from a Cox proportional hazards model with the estimated exposure propensity scores as a continuous variable and previous use of DMARD and/or corticosteroids. These drugs included cytotoxic DMARDs (other than MTX), noncytotoxic DMARDs, and corticosteroids. The analyses for subjects age 65 years or older in the 3 regions were combined in a stratified Cox proportional hazards regression to allow for different baseline cancer incidence among the 3 regions.

RESULTS

Characteristics of the study patients.

After the application of exclusion and inclusion criteria, the pooled cohort from the 3 regions consisted of 29,422 eligible RA patients. We restricted the analysis to 7,830 subjects who had used a biologic DMARD and/or MTX. Of 1,152 biologic DMARD users (etanercept in 743 [64%], infliximab in 381 [33%], and anakinra in 28 [2%]), 628 (55%) had used MTX previously and 451 (39%) were receiving MTX at the time of initiation of treatment with a biologic agent.

The characteristics of the study population ages 65 years and older, measured during the 6-month period before exposure to either a biologic DMARD or MTX, are shown in Table 1. The characteristics of the subjects who were originally identified as MTX users but later started treatment with a biologic DMARD were measured twice (once before initiation of MTX use and again before initiation of biologic DMARD use). Although there was some heterogeneity in their characteristics, possibly reflecting different patient populations and physician practices, biologic DMARD users generally had more severe RA than MTX users, as indicated by the percentage who had had RA-related surgery, extraarticular manifestations, a C-reactive protein test ordered, arthrocentesis, and use of other cytotoxic DMARDs. Markers of comorbidity (23), such as the number of physician visits, distinct number of generic medications, and Charlson index (24), were also higher in biologic DMARD users compared with MTX users. C-statistics of propensity score models at baseline were 0.79 for New Jersey, 0.74 for Pennsylvania, and 0.81 for British Columbia, indicating good discrimination.

Table 1. Characteristics of RA patients in the biologic DMARD treatment group and the MTX treatment group age ≥65 years, by region*

Variable	Pennsylvania		New Jersey		British Columbia		Total
	Biologic DMARD (n = 462)	MTX (n = 2,365)	Biologic DMARD (n = 443)	MTX (n = 2,029)	Biologic DMARD (n = 247)	MTX (n = 2,912)	Biologic DMARD (n = 1,152)	MTX (n = 7,306)
* Variables were measured during the 6 months prior to initiation of the study drug. On average, 55% of the patients in the biologic disease-modifying antirheumatic drug (DMARD) group were also included in the methotrexate (MTX) group because their treatment was switched from MTX to a biologic DMARD during the study period. These patients' characteristics were measured during 2 different time frames (the 6 months prior to initiation of treatment with each drug). Except where indicated otherwise, values are the percent of patients. RA = rheumatoid arthritis; CRP = C-reactive protein; ESR = erythrocyte sedimentation rate; RF = rheumatoid factor. † Data not available for the British Columbia cohort. ‡ Azathioprine, cyclosporine, and leflunomide. § Sulfasalazine, hydroxychloroquine, auranofin, gold sodium thiomalate, D-penicillamine, and minocycline.
Demographic characteristics
Age, mean ± SD years	71.7 ± 5.6	73.9 ± 6.3	70.6 ± 5.4	73.0 ± 6.4	72.2 ± 4.9	73.1 ± 5.8	71.4 ± 5.4	73.4 ± 6.2
Female	90.7	89.3	91.0	89.1	75.3	73.1	75.3	73.1
White	93.4	93.8	79.5	81.1	†	†	†	†
Health service utilization
Charlson comorbidity score, mean ± SD	2.0 ± 1.5	1.8 ± 1.2	2.1 ± 1.5	1.8 ± 1.3	1.9 ± 3.2	0.7 ± 1.6	2.0 ± 2.0	1.4 ± 1.5
No. of physician visits, mean ± SD	7.7 ± 5.2	2.5 ± 4.0	7.7 ± 5.7	2.7 ± 4.3	20.1 ± 10.7	13.9 ± 10.3	10.5 ± 8.7	7.1 ± 9.1
No. of medicines taken, mean ± SD	9.0 ± 5.2	6.7 ± 3.9	10.2 ± 6.2	6.9 ± 5.3	4.3 ± 4.5	3.1 ± 2.8	8.5 ± 5.9	5.3 ± 4.4
Any hospitalization	19.7	19.2	19.4	16.1	6.1	5.7	16.7	13.0
RA severity–related covariates
RA-related surgery	3.0	1.7	2.5	1.7	4.9	2.7	4.9	2.7
Extraarticular manifestations	25.5	9.7	23.3	9.1	3.6	4.0	20.0	7.2
Difficulty walking	4.3	4.1	11.1	11.4	0.0	0.0	6.0	4.5
Pain in joints	32.5	27.0	28.2	25.4	8.5	10.5	25.7	20.0
CRP test ordered	16.9	8.9	22.1	12.9	21.1	8.5	19.8	9.8
ESR test ordered	37.7	39.4	48.3	49.4	55.5	54.2	45.6	48.1
RF test ordered	8.9	12.1	17.4	23.5	7.3	24.6	11.8	20.3
Arthocentesis	34.2	30.5	30.5	23.9	2.0	1.1	25.9	17.0
Injection (joint, tendon, etc.)	2.2	2.4	10.4	7.3	10.9	8.1	7.2	6.0
Corticosteroid treatment	41.1	23.9	51.7	46.2	69.6	52.4	51.3	41.5
Other cytotoxic DMARD treatment‡	12.8	2.2	18.1	2.7	47.4	1.9	22.2	2.2
Noncytotoxic DMARD treatment§	18.4	20.0	20.1	18.1	36.4	31.5	22.9	24.1

Cancer incidence rates.

Table 2 shows the number of cancers, person-years of followup, and incidence rates of specific cancers examined in the study. We observed heterogeneity among regions and variability due to the small number of cases. Table 3 summarizes the number of observed and expected cancers and standardized incidence ratios (SIRs) for each cancer category among the study population (age ≥65 years) compared with the general population. The expected number of cancers was calculated using age- and sex-specific incidence rates for the general population from Surveillance Epidemiology and End Results data (25). The SIRs for non-Hodgkin's lymphoma and multiple myeloma were more than double as compared with the general population, which is compatible with previous findings (1–10). We also found significantly elevated risks of melanoma and lung, colorectal, and urinary tract/bladder cancers in the study population compared with the general population.

Table 2. Number of cases of cancer by type, person-years studied, and incidence rates in the Pennsylvania, New Jersey, and British Columbia cohorts*

Cancer type	Biologic DMARD group			MTX group
	No. of cases	Person-years	Incidence rate	No. of cases	Person-years	Incidence rate
* Incidence rates are per 100,000. DMARD = disease-modifying antirheumatic drug; MTX = methotrexate.
Non-Hodgkin's lymphoma
PA	3	1,139.5	263.3	17	10,488.6	162.1
NJ	1	1,413.4	70.8	17	9,351.7	181.8
BC	0	400.6	0.0	20	10,541.2	189.7
Multiple myeloma
PA	1	1,139.4	87.8	5	10,516.3	47.5
NJ	3	1,414.6	212.1	5	9,395.8	53.2
BC	0	400.6	0.0	5	10,543.8	47.4
Leukemia
PA	1	1,142.8	87.5	4	10,520.7	38.0
NJ	1	1,410.3	70.9	5	9,406.1	53.2
BC	1	400.5	249.7	10	10,543.3	94.8
Melanoma
PA	1	1,138.5	87.8	5	10,511.0	47.6
NJ	0	1,412.9	0.0	15	9,370.9	160.1
BC	0	400.6	0.0	8	10,543.8	75.9
Colorectal cancer
PA	6	1,121.2	535.1	33	10,447.8	315.9
NJ	2	1,398.9	143.0	44	9,298.3	473.2
BC	1	400.5	400.6	32	10,538.6	303.6
Lung cancer
PA	0	1,143.2	0.0	47	10,490.7	448.0
NJ	14	1,404.2	997.0	49	9,364.0	523.3
BC	3	399.5	750.9	56	10,534.9	531.6
Gastric cancer
PA	0	1,143.2	0.0	4	10,516.8	38.0
NJ	0	1,415.9	0.0	6	9,393.3	63.9
BC	0	400.6	0.0	2	10,544.3	19.0
Breast cancer
PA	0	1,032.5	0.0	34	9,487.1	358.4
NJ	6	1,297.0	462.6	32	8,313.6	384.9
BC	0	315.4	0.0	40	7,773.5	514.6
Prostate cancer
PA	0	102.8	0.0	13	903.1	1,439.5
NJ	0	99.5	0.0	12	943.0	1,272.5
BC	1	84.9	1,177.2	28	2,754.1	1,016.7
Uterine corpus/cervical cancer
PA	0	1,040.3	0.0	8	9,555.9	83.7
NJ	1	1,309.3	76.4	3	8,428.6	35.6
BC	0	315.4	0.0	4	7,782.5	51.4
Ovarian cancer
PA	0	1,040.3	0.0	6	9,582.4	62.6
NJ	1	1,308.8	76.4	10	8,425.3	118.7
BC	0	315.4	0.0	4	7,782.9	51.4
Urinary tract/bladder cancer
PA	3	1,134.3	264.5	16	10,499.2	152.4
NJ	4	1,406.3	284.4	14	9,385.0	149.2
BC	0	400.6	0.0	17	10,541.6	161.3
Renal cancer
PA	0	1,143.2	0.0	6	10,518.9	57.0
NJ	3	1,408.5	213.0	6	9,407.0	63.8
BC	0	400.6	0.0	4	10,544.5	37.9

Table 3. Person-years, observed number of cancers by type, expected number of cancers by type, and standardized incidence ratios for the study population compared with the US general population age ≥65 years*

Cancer type	Person-years	Observed	Expected†	SIR	95% CI
* SIR = standardized incidence ratio; 95% CI = 95% confidence interval. † Calculated using age- and sex-specific incidence rates in the general population, from the National Cancer Institute's Surveillance Epidemiology and End Results (25).
Non-Hodgkin's lymphoma	33,335.0	58	26.0	2.2	1.71–2.87
Multiple myeloma	33,410.0	19	9.3	2.0	1.26–3.12
Leukemia	33,423.7	22	16.6	1.3	0.85–1.97
Melanoma	33,377.7	29	12.8	2.3	1.55–3.22
Colorectal cancer	32,844.9	118	97.3	1.2	1.01–1.45
Lung cancer	31,532.8	169	95.6	1.8	1.52–2.05
Gastric cancer	33,414.1	12	14.2	0.8	0.46–1.44
Breast cancer	28,219.1	112	126.6	0.9	0.73–1.06
Prostate cancer	4,887.4	54	51.6	1.0	0.79–1.35
Uterine corpus/cervical cancer	28,432.0	16	30.9	0.5	0.31–0.82
Ovarian cancer	28,455.1	21	15.1	1.4	0.88–2.09
Urinary tract/bladder cancer	33,367.0	54	26.4	2.0	1.55–2.65
Renal cancer	33,422.7	19	12.9	1.5	0.91–2.25

Risk of cancer associated with biologic DMARD use.

Table 4 shows pooled HRs for lymphoproliferative, hematologic, solid, and overall cancers. We found no significant increase in risk of cancers in biologic DMARD users versus MTX users.

Table 4. Pooled effects of biologic DMARDs, compared with MTX, on cancer in patients age ≥65 years*

Adjustment	Cancer type
	Lymphoproliferative		Hematologic		Solid		Overall
	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
* Lymphoproliferative cancers include non-Hodgkin's lymphoma, chronic lymphocytic leukemia, and multiple myeloma. Hematologic cancers include leukemia, lymphoma, and multiple myeloma. Solid cancers include colorectal, lung, stomach, breast, prostate, uterine, ovarian, urinary tract/bladder, and renal cancers and melanoma. Overall cancers include hematologic or solid cancers. DMARDs = disease-modifying antirheumatic drugs; MTX = methotrexate; HR = hazard ratio; 95% CI = 95% confidence interval. † Stratified Cox proportional hazards model by region, with study time as a time scale. ‡ Stratified Cox proportional hazards model by region, with study time as a time scale and age and sex included in the model. § Stratified Cox proportional hazards model by region, with study time as a time scale and adjustment for rheumatoid arthritis drug use and propensity score as a continuous variable.
Unadjusted†	1.20	(0.57–2.51)	1.45	(0.76–2.74)	0.91	(0.66–1.25)	1.00	(0.75–1.33)
Age/sex-adjusted‡	1.18	(0.56–2.48)	1.44	(0.76–2.72)	0.92	(0.67–1.26)	1.00	(0.75–1.34)
Fully adjusted§	1.11	(0.51–2.37)	1.37	(0.71–2.65)	0.91	(0.65–1.26)	0.98	(0.73–1.31)

In a secondary analysis, we assumed a 180-day induction period and excluded the first 180 days of followup and events occurring during this period. Pooled adjusted HRs for biologic DMARD users were 0.92 (95% confidence interval [95% CI] 0.39–2.20) for lymphoproliferative disorders, 1.11 (95% CI 0.52–2.38) for hematologic malignancies, 0.98 (95% CI 0.69–1.38) for solid tumors, and 0.99 (95% CI 0.71–1.36) for overall cancers, similar to the results obtained in the primary analysis assuming no induction time. In another secondary analysis we estimated the effect of TNFα antagonists only, excluding interleukin-1 receptor blockade (anakinra) treatment; results were similar to those in the primary analysis.

We tested the sensitivity of our results to different claims-based definitions of cancers. An alternative definition, also analyzed, was 2 diagnoses of cancer in 2 months, which was shown in a validation study to have higher specificity and slightly lower sensitivity. With the alternative definition, pooled adjusted HRs were 1.12 (95% CI 0.52–2.42) for lymphoproliferative disorders, 1.33 (95% CI 0.69–2.58) for hematologic malignancies, 0.97 (95% CI 0.69–1.35) for solid tumors, and 1.01 (95% CI 0.75–1.37) for overall cancers. The differences in point estimates for solid and overall cancers did not exceed 5%, and 95% CIs for all estimates overlapped widely, with no systematic differences.

DISCUSSION

We combined a large health care utilization database from 3 regions in the US and Canada and estimated the risk of cancers among users of biologic DMARDs, including anti-TNFα antagonists, compared with users of MTX. We found no significant increase in the risk of cancers in biologic DMARD users. Our data indicate that it is unlikely that RA patients who have received biologic agents have a much greater risk of lymphoproliferative disorders, hematologic malignancies, and solid tumors as compared with MTX users. The strengths of our study include the use of a large combined data set with high-risk elderly populations and an internal comparison group with RA. To reduce confounding by RA severity, we used time-varying propensity score adjustment. We also chose MTX users as the comparison group because RA severity in patients treated with this drug is more similar than in those treated with other DMARDs.

Our results do not rule out an increased risk of cancers in RA patients treated with biologic DMARDs compared with those receiving non-MTX DMARDs or with non-RA populations. However, a prospective cohort study in France that identified all cases of lymphoma in RA patients treated with MTX over a 3-year period (1996–1998) showed that the risk of non-Hodgkin's lymphoma was not significantly elevated in MTX users versus the general population (26).

Wolfe and Michaud determined lymphoma incidence from followup of 18,572 patients by 908 US rheumatologists from 1998 to 2002 and compared incidence by treatment group with that in the general population (16). Age/sex-adjusted SIRs compared with the general population were 2.6 for patients treated with infliximab, 3.8 for patients treated with etanercept, 1.7 for patients treated with MTX alone, and 1 in patients not treated with MTX or a biologic DMARD. Comparative rate ratios (27) for infliximab and etanercept versus MTX calculated from the study were 2.3 and 4.0, respectively. The apparent increased cancer risk among users of biologic agents in the study by Wolfe and Michaud could be explained by the difference in the underlying risk among their study patients, who were treated exclusively by rheumatologists, compared with our study patients, many of whom were not treated by rheumatologists. Additionally, Wolfe and Michaud did not adjust for patient characteristics other than age and sex, whereas our models included adjustment for a broader range of variables.

Franklin and colleagues used a Swedish registry of patients treated with TNFα antagonists and community-based RA patients treated with other DMARDs to examine the risks of various cancers in TNFα antagonist users (18). Sixteen cancers (5 lymphomas) per 1,603 person-years were identified in the TNFα antagonist group and 69 cancers (2 lymphomas) per 3,948 person-years were identified in the comparison group. The adjusted HR for lymphoma was 4.9 in anti-TNFα–treated patients compared with patients who had never received anti-TNFα, suggesting a large increase in risk in the TNFα antagonist–treated group. This increased HR remained after adjusting for 1 patient-reported marker of RA severity (patient-reported Health Assessment Questionnaire scores [28]).

Although we cannot extrapolate the risk among biologic DMARD users as compared with MTX users from the data presented in the Swedish study, the difference between the findings of that study and our results is likely due to different comparison groups. The comparison group in the Swedish study, RA patients who were never treated with TNFα antagonists although they may have been treated with other DMARDs, could have had much less severe RA than the TNFα antagonist–treated patients. Therefore, it is more difficult to distinguish a treatment effect from the effect of RA severity on cancer incidence. In more recent studies from Sweden (29, 30), which included the patients in Franklin and colleagues' study, results were similar to ours.

A previous study with detailed clinical data showed that high RA inflammatory activity was associated with a 26-fold increased risk of lymphoma compared with low activity (31). RA severity and activity influence the choice of therapy and were therefore considered the strongest potential confounders in our study. To control for RA severity, we first chose MTX users for comparison because we would expect the severity of RA to be similar in biologic DMARD–treated and MTX-treated groups. Although our data set does not have precise information on RA severity, such as inflammatory activity or functional capacity, we adjusted for multiple factors in the claims data that were considered to be related to RA severity and reduced them into a time-varying propensity score.

The propensity score model had good discriminating power to differentiate biologic DMARD users from MTX users, but the adjustment by propensity score and other important factors such as age, sex, and/or RA drug use did not change effect estimates significantly, suggesting that RA severity was similar in the biologic DMARD group and the MTX group. An alternative explanation for this could be that little confounding by severity was captured by our claims data. However, Johannes et al validated their claim-based propensity score predicting etanercept use versus use of other DMARDs by comparing medical record–derived RA severity indicators (32) and found that, after adjusting for claim-based propensity scores, the RA severity indicators, e.g., swollen joints and morning stiffness, were similar between patients treated with etanercept and those treated with other DMARDs.

Our claims-based definitions of cancers were shown to have very high specificity in a separate validation study. We assessed the degree of potential misclassification bias introduced given the reported specificity and sensitivity. For lymphoproliferative malignancies, the observed HR of 1.11 changed to 1.20 after correcting the estimates with the reported specificity and sensitivity. This demonstrated that our claims-based definition of cancers was unlikely to introduce meaningful bias.

Our data indicate that, in patients with RA, biologic DMARD treatment is unlikely to be associated with increased risk of lymphoproliferative disorders, hematologic malignancies, solid tumors, or overall cancers compared with MTX treatment, although a small increase or decrease in such risk is still possible. Even with the use of a large combined data set, it is a challenge to investigate the effect of a relatively rare exposure (biologic DMARDs) on rare diseases (hematologic malignancies). Larger collaborative studies and/or longer followup will be needed in order to obtain more precise estimates.

Acknowledgements

The authors thank Jeffrey Peppercorn, MD, MPH (University of North Carolina at Chapel Hill) for useful comments on the claims data–based definitions for incident cancers and on the draft version of this work, and Raisa Levin, MS, Claire F. Canning, MS, and Helen Mogun, MS for their contributions in extracting and analyzing the data.

APPENDIX A

ESTIMATION OF TIME-VARYING PROPENSITY SCORE

We used a 2-step approach to estimate the propensity score (PS), i.e., the probability of receiving a biologic DMARD versus MTX given measured covariates. First, we calculated a baseline PS model at the time patients entered the study cohort. From the list of variables tested as potential covariates (available upon request from the corresponding author), we identified 80 patient characteristics measurable in our databases that were potential confounders based on a priori knowledge or that were variables with some explanatory power for the DMARD exposure status, indicated by c-statistics of >0.51 (33, 34). These characteristics were used as independent covariates in a logistic regression model of current DMARD exposure. We used a backward selection algorithm to derive the baseline PS model (P ≤ 0.15 to remain). This is a pragmatic approach, although a recent study showed that it might not yield the most efficient PS (35).

In a second step, we created a time-varying PS. All covariates in the baseline PS model were measured repeatedly every 6 months. Generalized estimating equations with an exchangeable correlation structure were used to fit a logistic regression model of treatment status as a function of repeatedly measured PS covariates. This yielded the final coefficient estimates used to calculate PS for each study subject every 6 months.

Because it was impossible for subjects to be receiving a biologic DMARD before these agents became available, estimation of propensity scores was limited to patient-time following the market introduction of biologic DMARDs (August 24, 1998 for the US data and March 1, 2001 for the Canada data). We used the PS parameter estimate derived after the marketing of biologic DMARDs, which is interpreted as the propensity of receiving a biologic DMARD had they been marketed, to calculate time-varying PS for patient-time prior to the marketing of these agents. Patient-time prior to marketing of biologic DMARDs was included in our analysis to increase the statistical power of the study by increasing patient-time receiving nonbiologic DMARDs.