Hematologic Malignant Neoplasms After Drug Exposure in Rheumatoid Arthritis

Background  Rheumatoid arthritis is a severe inflammatory polyarthritis that requires long-term treatment with disease-modifying antirheumatic drugs. There is increasing concern about the influence of rheumatoid arthritis therapy on the risk for hematologic malignant neoplasms.

Methods  We used a case-control design nested in a cohort of 23 810 patients with rheumatoid arthritis assembled from administrative databases covering the population of Quebec, Canada. The study was carried out from January 1, 1980, through December 31, 2003. Case patients having hematologic malignant neoplasms were ascertained from physician billing and hospitalization records; each case patient was matched for age and sex with 10 control subjects. Adjusting for clinical variables and concomitant medications, we used conditional logistic regression to analyze potential associations between disease-modifying antirheumatic drug exposures and risk for hematologic malignant neoplasms. We estimated rate ratios attributable to each disease-modifying antirheumatic drug exposure.

Results  During the study, hematologic malignant neoplasms developed in 619 patients, including lymphomas in 346 patients, leukemia in 178 patients, and multiple myelomas in 95 patients. The unadjusted rate ratios for hematologic malignant neoplasms after drug exposures were as follows: methotrexate, 1.18 (95% confidence interval [CI], 0.99-1.40); azathioprine, 1.44 (95% CI, 1.01-2.03); and cyclophosphamide, 2.21 (95% CI, 1.52-3.20). Because biologic agents first appeared in the Régie d’Assurance Maladie du Quebec formulary in 2002, there were few exposures to these drugs. Adjusted estimates suggested that hematologic cancer risk was most elevated after exposure to cyclophosphamide (rate ratio, 1.84; 95% CI, 1.24-2.73). For lymphomas only, the adjusted rate ratio after cyclophosphamide exposure was 2.12 (95% CI, 1.33-3.54).

Conclusions  In this large cohort of patients with rheumatoid arthritis, the greatest relative risk for hematologic malignant neoplasms was noted after use of cyclophosphamide. Assessments of risk related to newer and emerging therapies should carefully consider previous and concomitant medication exposures.

Rheumatoid arthritis (RA) is a chronic, often disabling inflammatory polyarthritis affecting up to 1% of the adult population. The etiology of this autoimmune condition is incompletely understood, but joint inflammation and damage may be curtailed by the use of disease-modifying antirheumatic drugs (DMARDs) that target the immune system. Increasingly over the last 3 decades, data support an important increased risk for hematologic malignant neoplasms in RA, particularly for lymphoma,^1,2 although development of leukemia is also of concern. It has been proposed that DMARDs could increase hematologic malignant neoplasm risk by either direct gene mutation or disturbing immune surveillance, thus preventing the deletion of abnormal cells.

The issue is of increasing interest given reports of lymphoma in patients receiving newer therapies such as anti–tumor necrosis factor (TNF) agents. One of the difficulties in assessing the effects of these newer agents is the need for precise information about risk related to the use of traditional DMARDs, which are often used either successively or in combination with anti-TNF agents. The purpose of our study was to assess the risk in RA of hematologic malignant neoplasms associated with DMARD exposures.

Our cohort with RA was assembled using administrative data from the Régie d’Assurance Maladie du Québec (RAMQ) physician billing database and the Ministry of Health's Maintenance et Exploitation des Données Pour l’Etude de la Clientèle Hospitalière (MEDECHO) hospital discharge database. For residents of the province of Quebec, health service coverage is universal. The RAMQ database documents physician services, both inpatient and outpatient, for all provincial beneficiaries, and the MEDECHO hospital discharge database captures dates and discharge diagnoses, as provided by the International Classification of Diseases, Ninth Revision [ICD-9]. The discharge diagnoses are assigned by trained medical coders who systematically read through each medical record to identify up to 15 medical conditions, using a tool for mapping ICD-9 codes.

For our analyses, the subject pool was restricted to those individuals registered in the RAMQ pharmacy claims database, which includes provincial residents 65 years or older, those who receive social assistance, and those who do not have private drug insurance. This encompasses about half of all Quebec residents, that is, essentially all residents 65 years or older. The RAMQ and MEDECHO databases are linkable through the personal health insurance number, a 10-digit unique identifier for each beneficiary.

Between January 1, 1980, and December 31, 2003, we identified all individuals having a diagnosis of RA, using physician billing activity (ICD-9 code 714). Subjects were required to have at least 1 ICD-9 code for RA in the billing data plus at least 1 DMARD exposure after January 1, 1980. Cohort entry was defined by the date of the first DMARD prescription. These included methotrexate, hydroxychloroquine sulfate, chloroquine, sulfasalazine, azathioprine, leflunomide, cyclophosphamide, cyclosporine, gold compounds, minocycline, penicillamine, and anti-TNF agents. Anti-TNF agents were included in the provincial formulary only as of 2002, and no other biologic agents were available during the study.

Subjects were required to have more than 3 months of eligibility in the health insurance plan before cohort entry and to have no record of hematologic malignant neoplasms before their cohort entry date, as determined by ICD-9 billing or hospitalization discharge diagnoses. We observed the subjects from cohort entry to the first of 3 possible events: the outcome of interest, death, or the end of the study (December 31, 2003). The outcome of interest was any new hematologic malignant neoplasm, defined by the occurrence of an outpatient visit or hospital discharge in which the primary diagnosis was one of the relevant ICD-9 codes (ie, codes 200-208). This approach to cancer ascertainment has been validated in other studies.^3,4

To most efficiently handle the complexity of the time-dependent medication exposures, we used a nested case-control design.⁵ For each new case of a hematologic malignant neoplasm occurring in our RA cohort, we randomly selected 10 control subjects with RA, matching for age, sex, and month and year of cohort entry and ensuring that each control subject was cancer free (ie, had no record of malignant neoplasms as determined by ICD-9 billing or hospitalization discharge diagnoses) on the day the case occurred (this was the index date for each case-control set). Medication use anytime from cohort entry to the index date was identified from the RAMQ prescription database.

Disease severity was not directly measured, but we adjusted for correlates of disease severity occurring before the index date: glucocorticoid use, number of all physician visits including visits for reasons other than RA, and extra-articular RA features. Controlling for glucocorticoid use and number of physician visits as an approach to adjust for RA severity has been successfully applied in previous studies.⁶ In addition, extra-articular RA features, such as rheumatoid lung, Felty syndrome, neuropathy, and vasculitis, are good correlates of RA severity.^7-9 These were ascertained in our study based on ICD-9 code diagnoses from inpatient or outpatient physician visits.

We calculated person-years of follow-up to determine the incidence rate for hematologic malignant neoplasms in the RA cohort. Conditional logistic regression was performed on the nested case-control sample to estimate the rate ratio (RR) of hematologic malignant neoplasm occurrence for each DMARD, along with 95% confidence intervals (CIs) for the adjusted estimates. The DMARDs were analyzed in terms of time-dependent use, assessing the independent effects of the most commonly used types. The adjusted RR estimates reflect the specific effect of each exposure, adjusted for whether subjects had received other medications. All DMARDs were included in models that controlled concurrently for other medications, number of physician visits, and extra-articular RA features.

We identified a cohort of 23 733 patients with RA with DMARD exposures and no recorded history of cancer. At cohort entry, mean (SD) age was 61.7 (14.6) years. Most subjects (70.1%) were women, as expected in RA. The most common current DMARD exposures at cohort entry were methotrexate, antimalarial agents, and sulfasalazine. Subjects were followed up on average for a mean (SD) of 6.7 (5.1) years. This yielded a total of 158 067 person-years of follow-up, during which 619 hematologic malignant neoplasms developed, for an incidence rate of 391.6 cases per 100 000 person-years.

The demographics for the 619 case patients vs their 6190 matched control subjects were as follows: mean (SD) age, 70 .0 (12.4) years vs 70.2 (11.4) years; female sex, 65.4% for both cohorts; and mean (SD) number of physician visits during follow-up, 26.3 (26.1) vs 18.1 (14.2). The events occurred at a mean (SD) of 5.2 (3.9) years after cohort entry. There was a trend for a higher number of physician visits (calculated up to the index date only) for the cases compared with the controls, but variance was high. The prevalence of many extra-articular RA features (eg, vasculitis and scleritis) seemed similar among cases and controls, although the prevalence of Felty syndrome was more common among cases compared with controls (difference of 8.9%; 95% CI, 6.8-11.5). Neuropathy was also slightly more common among cases (difference of 1.7%; 95% CI, 0.8-3.1).

At univariate analyses of the risk for hematologic malignant neoplasms, there was a statistically significant association only with azathioprine (RR, 1.44; 95% CI, 1.01-2.03) and cyclophosphamide (RR, 2.21; 95% CI, 1.52-3.20) therapies. The adjusted RR estimates reflecting the specific effect of each exposure, independent of whether the subjects were concomitantly receiving the other medications, are given in the Table. As indicated by these estimates, the risk seemed to be most elevated with exposures to cyclophosphamide (RR, 1.84; 95% CI, 1.24-2.73). For anti-TNF agents, the confidence intervals were relatively wide, although the RR point estimate was elevated, in both univariate and adjusted estimates.

The most frequently noted specific types of hematologic malignant neoplasms were lymphoma (n = 346), leukemia (n = 178 cases), and multiple myeloma (n = 95). For lymphoma only, the adjusted RR after methotrexate exposure was 1.23 (95% CI, 0.97-1.57); after azathioprine exposure, 1.09 (95% CI, 0.67-1.77); after cyclophosphamide exposure, 2.12 (95% CI, 1.33-3.54); and after anti-TNF exposure, 3.14 (0.58-17.1). Sensitivity analyses using cumulative exposures and 5-year intervals between exposures did not change the RRs appreciably.

There have been a handful of controlled studies evaluating DMARD use and risk for malignant neoplasm. Asten et al¹⁰ assessed cumulative exposures to these in a large (N = 17 773) multicenter cohort of patients with rheumatic disease and found an increased risk for hematologic malignant neoplasms; however, disease severity was not controlled for. Two early case-control studies of the risk of hematologic malignant neoplasms in RA^11,12 suggested a link with azathioprine and cyclophosphamide therapies, although not definitively. More recently, Baecklund et al¹³ showed an association when controlled for disease activity of hematologic malignant neoplasms with azathioprine use in RA.

Wolfe and Michaud,² in their large clinic-based RA sample, estimated an approximate 3-fold increased risk for lymphoma in patients after anti-TNF exposure and no definite effects for methotrexate, compared with those not receiving either agent. In that study, there were insufficient data to completely explore disease severity. Geborek et al¹⁴ compared cancer risk in 1557 patients with RA treated with DMARDs, making some adjustment for disease severity. They reported no increased risk of cancer in RA overall, although they did note increased risk for lymphoma in those receiving anti-TNF agents compared with those receiving traditional DMARDs. After adjustment for baseline disease severity, the risk for lymphoma developing in the anti-TNF–treated patients was 5 times the risk in unexposed patients, though with a wide CI, making definitive conclusions impossible.

For a rare disease such as RA and a relatively infrequent outcome, studies using administrative databases can be efficient. However, this approach does not allow confirmation of the RA diagnosis or the outcome. Insofar as the RA diagnosis, combining billing codes with DMARD prescription data enhances the validity of the RA diagnosis, as has been demonstrated by other investigators.^15,16 Some reassurance is provided in that the demographics for our cohort (ie, age and sex) were similar to those in clinical RA populations. For cancers identified through administrative data, though we cannot confirm these per se, we emphasize that discharge diagnoses in Canadian hospitalization databases are assigned by trained medical coders; validation work has shown that the discharge databases can be nearly as reliable as medical records.¹⁷ Hospital discharge diagnoses are the basis for the provincial tumor registry in Quebec. Similarly, as for physician billing data, some validation work has suggested that billing records in Quebec are valid and more accurate than patient self-reports.¹⁸ For these reasons, we believe our approach to the identification of malignant neoplasms is reasonable. However, as in all observational studies, caution must be used in interpretation of the results.

In pharmacoepidemiology, bias may arise if medication use occurs differentially according to preexisting risk for the outcome (channeling). We attempted to control for this using appropriate strategies. We purposely created a homogeneous population with respect to the likelihood of drug exposure. This was done by assembling an RA cohort whose members all were exposed to DMARDs and by examining risk for DMARD type. In addition, we excluded subjects with a history of cancer. We also adjusted for correlates of RA disease severity, including glucocorticoid use, number of physician visits, and extra-articular RA features. Using glucocorticoid exposure as a surrogate of RA disease severity is not without problems. Initiation of steroid treatment may correlate with disease severity, but prolonged treatment could lead to better-controlled disease activity than in patients not treated with steroids.

Our adjusted analyses reflect the specific independent effects of each DMARD, adjusted for other DMARD exposures. Our cohort consisted of both prevalent and incident RA cases; thus, some medication exposures, if occurring before 1980, may not have been recorded. However, we do not consider this a major limitation of our study because widespread DMARD use began primarily in the 1980s.

For severe refractory RA, oral cyclophosphamide was a treatment option even up to the last decade.¹⁹ In our cohort, the patients exposed to cyclophosphamide, compared with those not exposed to cyclophosphamide, demonstrated a greater prevalence of Felty syndrome (19.4% vs 6.4%; difference of 12.7% [95% CI for difference, 7.6-18.9]) and vasculitis (14.6% vs 0.4%; difference of 14.2% [95% CI for difference, 9.8-19.9]). Other types of extra-articular RA were similar between patients exposed to cyclophosphamide compared with those not exposed.

Because lymphoma is a cancer arising from lymphocytes, it is not surprising that diseases characterized by immune system dysregulation (including RA, among others) are associated with malignant transformation of lymphocytes. Similarly, T-cell large granular lymphocyte leukemia is a disorder often associated with autoimmune disorders, especially RA, and particularly in patients with Felty syndrome. This malignancy of cytotoxic T cells is characterized by dysregulated apoptosis and may be antigen driven.²⁰ Controversy continues, however, as to how much of the total risk of hematologic malignant neoplasms in RA is related to the disease process itself vs to immunosuppressive medications; recent data suggest both aspects are likely important.¹³ Immunosuppressive drugs have been linked to the development of both leukemia and lymphomas, and immunodeficiency, whether innate or acquired, is a strong risk factor for certain lymphomas. In many cases, this association relates to the emergence of Epstein-Barr virus–driven malignant proliferations; however, in RA, this does not seem to be the only driving force.¹³ Impaired immune surveillance can prevent the deletion of abnormal cells, a phenomenon that may have been magnified in our study design, which focused on an older cohort.

In conclusion, in our large RA cohort, increased risk for hematologic malignant neoplasms was most evident with cyclophosphamide exposure. This emphasizes the need, when evaluating malignancy risk related to novel agents, for careful consideration of previous and concomitant medication exposures.

Correspondence: Sasha Bernatsky, MD, PhD, Division of Clinical Epidemiology, McGill University Health Centre, 687 Pine Ave W, V Bldg, Montreal, QC H3A 1A1, Canada.

Accepted for Publication: September 17, 2007.

Author Contributions:Study concept and design: Bernatsky and Suissa. Acquisition of data: Bernatsky and Suissa. Analysis and interpretation of data: Bernatsky, Clarke, and Suissa. Drafting of the manuscript: Bernatsky. Critical revision of the manuscript for important intellectual content: Bernatsky, Clarke, and Suissa. Statistical analysis: Suissa. Obtained funding: Suissa. Administrative, technical, and material support: Suissa. Study supervision: Clarke and Suissa.

Financial Disclosure: None reported.

Funding/Support: This study was supported by a Canadian Institutes for Health Research operating grant. Dr Bernatsky is the recipient of Career Awards from the Canadian Institutes for Health Research (CIHR), Fonds de Recherche en Santé du Québec (FRSQ), and the Canadian Arthritis Network, and is supported by the McGill University Research Institute and Faculty of Medicine. Dr Clarke is an FRSQ National Scholar. Dr Suissa is a recipient of the James McGill Professorship and the CIHR Distinguished Investigator Award.