Spontaneous reports of interstitial lung disease (ILD) in patients with rheumatoid arthritis (RA) treated with leflunomide, a disease-modifying antirheumatic drug (DMARD), have been appearing recently. To assess this risk, we conducted a population-based epidemiologic study.
A cohort of 62,734 patients with RA to whom a DMARD had been dispensed between September 1, 1998 and December 31, 2003 was formed using the PharMetrics claims database. A nested case–control design was used, in which each case of serious ILD requiring hospitalization was matched to 100 controls according to age (calendar time) and equal or greater duration of followup, to estimate adjusted rate ratios (RRs) of serious ILD associated with DMARD use.
There were 74 cases of serious ILD, which corresponds to a rate of 8.1 per 10,000 patients per year. The risk of ILD was increased with the use of leflunomide (adjusted RR 1.9 [95% confidence interval (95% CI) 1.1–3.6]). Among subjects with no previous methotrexate use and no history of ILD, the risk associated with leflunomide treatment was not elevated (RR 1.2 [95% CI 0.4–3.1]), but it was elevated among the remaining subjects (RR 2.6 [95% CI 1.2–5.6]). Patients with a history of ILD were twice as likely to have been prescribed leflunomide as any other DMARD.
The reports of ILD associated with leflunomide use are likely the result of channeling of high-risk patients to leflunomide treatment, particularly those with a history of methotrexate use or preexisting ILD. Patients with no history of ILD and no previous methotrexate use show no excess risk of developing ILD with leflunomide treatment.
Leflunomide, a disease-modifying antirheumatic drug (DMARD), has been available in the US since 1998 for the treatment of active rheumatoid arthritis (RA) (1). Soon after its launch in Japan in 2003, 16 cases of interstitial pneumonitis were reported in that country, of which 5 resulted in death; several of the 16 patients had preexisting lung disease (2). A subsequent Japanese study reported 26 cases of leflunomide-associated acute lung injury (3). It is unclear whether leflunomide is directly related to the occurrence of such lung toxicity or whether other factors, including channeling bias, may explain this.
Commonly used DMARDs, in particular, methotrexate, have been associated with pulmonary toxicity (4, 5). Preexisting lung disease has also been reported to be a risk factor for the development of methotrexate pneumonitis (5). We thus conducted a large-scale cohort study to assess the risk of interstitial lung disease (ILD) associated with the use of leflunomide. We also assessed whether this risk is modified by a history of lung disease and of methotrexate use. Finally, we investigated the possibility of channeling bias by assessing whether patients with lung disease were more likely to be prescribed leflunomide when it became available.
SUBJECTS AND METHODS
Role of the funding source.
Sanofi-Aventis made all arrangements to acquire the databases examined in this study. The authors independently designed the study, analyzed the data, interpreted the results, and wrote the manuscript. The manuscript was reviewed and approved by Sanofi-Aventis prior to submission.
The study was performed using the PharMetrics Patient-Centric Database (6). This database consists of standardized information on claims data, including physician visits, dispensed prescription medications, and hospitalizations, from more than 75 different managed care organizations and currently encompasses data for more than 55 million unique patients. This database has been used previously to assess the hepatic and cardiovascular effects of DMARDs (7, 8).
We first identified the source population of all subjects with at least 1 diagnosis of RA (International Classification of Diseases, Ninth Revision [ICD-9] code 714) for an inpatient or an outpatient encounter between January 1995 and December 2003 (Figure 1).
We formed a DMARD-treated cohort that was defined as all subjects who received at least 1 prescription for a DMARD on or after September 1, 1998, the month leflunomide was approved in the US. The DMARDs included leflunomide, methotrexate, all biologic agents, gold compounds, antimalarial drugs, minocycline, penicillamine, sulfasalazine, and cytotoxic agents. Cohort entry was defined as the date of the first such DMARD prescription. To avoid excluding subjects with fewer than 12 months of eligibility in the health insurance plan prior to cohort entry, all such subjects had their cohort entry date moved forward in time and redefined as the first prescription for a DMARD after 1 year of eligibility in the database. Accordingly, subjects who after this 1-year shift forward had no prescriptions for a DMARD were excluded.
All subjects were tracked from cohort entry until the earliest of the following: the date of termination of enrollment in the health plan, the date of death, the end of the study period (December 31, 2003), or the date of the outcome of interest, namely, ILD (defined below). Subjects with the outcome of interest during the 1-year period prior to cohort entry were excluded. Subjects were 18 years of age or older at cohort entry.
To address channeling bias, we identified all subjects from the source population who received a prescription for a DMARD between September 1, 1998 and December 31, 2003. Among those who received leflunomide or a biologic agent during this period, the first prescription of either of these 2 drugs was taken as the index prescription. The remaining subjects were divided into 2 groups: those who took only methotrexate during the observation period and all other subjects. For these 2 groups, a DMARD prescription was selected at random among all DMARD prescriptions dispensed during the observation period and was taken as the index prescription. Subjects with fewer than 4 months of eligibility in the health plans prior to the index prescription were excluded.
Cases of probable drug-related ILD were identified from inpatient encounters as all subjects who were hospitalized with a first-time primary diagnosis of postinflammatory lung fibrosis (ICD-9 code 515), idiopathic fibrosing alveolitis (code 516.3), or other/unspecified alveolar pneumonopathies (codes 516.8 and 516.9). As a sensitivity analysis, we expanded the case definition to include ILD identified from outpatient encounters as well as mention of the relevant codes as secondary diagnoses during inpatient encounters.
Nested case–control design.
We used a nested case–control approach within the DMARD-treated cohort. This approach allows one to deal with the complexity in the patterns of multiple drugs used to treat RA and varying durations of treatment without significant loss of power (9).
For each case of ILD identified in the DMARD-treated cohort, we randomly selected 100 controls from the cohort after matching for age (within a year) and the date of cohort entry. Controls also had to be at risk, that is, still in the cohort and without a diagnosis of ILD, on the day the case occurred. This date was designated as the “index date” for both cases and their matched controls.
Measurement of exposure.
For the primary analysis, all anti-RA drugs received during the year prior to the index date for both cases and controls were identified from dispensed prescription data. The DMARDs were divided into 4 categories: methotrexate, leflunomide, biologic agents (namely, infliximab, etanercept, adalimumab, and anakinra), and traditional DMARDs (namely, antimalarial agents, sulfasalazine, gold salts, minocycline, penicillamine, azathioprine, cyclosporine, and other cytotoxic agents).
For the analysis addressing the channeling issue, the exposure of interest was the presence of an ILD diagnosis on any outpatient claim or as a secondary diagnosis on inpatient claims during the year prior to the index prescription.
The primary covariates included age, a matching factor, and sex, as well as all comorbid conditions based on diagnoses made during the year prior to the index date. These included cardiovascular disease, respiratory illness, diabetes mellitus, hypertension, hypercholesterolemia, cancer, gastrointestinal conditions, and central nervous system conditions. Other drugs prescribed for RA, such as glucocorticoids, nonsteroidal antiinflammatory drugs, and cyclooxygenase 2 inhibitors, were used as covariates.
Respiratory disease codes that might signal the presence of ILD prior to cohort entry was defined as any diagnosis of postinflammatory lung fibrosis (ICD-9 code 515), idiopathic fibrosing alveolitis (code 516.3), and other or unspecified alveolar pneumonopathies (codes 516.8 and 516.9) appearing on any outpatient claim or as a secondary diagnosis on inpatient claims during the year prior to cohort entry.
Conditional logistic regression was used for all analyses of the nested case–control study to estimate crude and adjusted odds ratios (ORs) of ILD associated with DMARD use during the year prior to the index date. Since these ORs are accurate estimators of the rate ratios (RRs), the latter term will be used throughout (10). All analyses were adjusted for the concurrent use of the other DMARDs, the other anti-RA drugs, as well as sex and comorbid conditions. To assess the effects of disease and treatment history on the RRs for leflunomide and methotrexate, we conducted analyses stratified by these factors.
To assess the question of channeling to leflunomide, we compared the subjects with an index DMARD prescription of leflunomide, a biologic agent, or a traditional DMARD, with those with an index DMARD prescription of methotrexate only. We used unconditional logistic regression to estimate the OR adjusted for the use of other anti-RA drugs and for comorbidity factors during the year prior to the index prescription.
The source population consisted of 235,272 subjects with a diagnosis of RA between September 1, 1998, and December 31, 2003. The DMARD-treated cohort consisted of 62,801 subjects, of which 67 were excluded because they had the outcome during the year prior to cohort entry (Figure 1). The resulting 62,734 subjects in the study cohort were, on average, 54 years of age at cohort entry, and 75% were women. The majority entered the cohort with a methotrexate or a traditional DMARD prescription (78%). There were 74 subjects hospitalized because of ILD during followup. Of these, 43 had a diagnosis of postinflammatory pulmonary fibrosis, 8 had a diagnosis of idiopathic fibrosing alveolitis, and 23 had other alveolar pneumonopathies. With almost 91,000 person-years of followup, the rate of hospitalizations because of ILD in this cohort was 8.1 per 10,000 per year. The case group had more prior comorbidity and included more men than the control group (Table 1).
Table 1. Baseline characteristics of cases hospitalized for interstitial lung disease (ILD) and their matched controls
Cases (n = 74)
Controls (n = 7,400)
Age, mean ± SD years
62 ± 12
61 ± 11
Followup, mean ± SD months
12 ± 10
12 ± 10
Sex, % women
Comorbidity in the year prior to cohort entry, %
Ischemic heart disease
Congestive heart failure
Other cardiovascular disease
Cardiovascular risk factors
Other concurrent diseases
Central nervous system conditions
ILD (outpatient or secondary inpatient diagnosis)
Other lung disease
The risk of ILD was increased with the use of leflunomide in the year prior to the index date (adjusted RR 1.9; 95% CI 1.1–3.6) (Table 2). The increase was less and was not significant with the use of methotrexate (RR 1.4; 95% CI 0.8–2.3). However, in stratified analyses, subjects who had not taken methotrexate during the year prior to cohort entry and who did not previously have ILD (37 cases and 4,259 controls) had an increased risk of ILD with methotrexate treatment (RR 3.1; 95% CI 1.5–6.4) but not with leflunomide treatment (RR 1.2; 95% CI 0.4–3.1). In contrast, among those who had previously taken methotrexate or who had a previous diagnosis of ILD (37 cases and 3,141 controls), the risk of ILD was elevated with leflunomide treatment (RR 2.6; 95% CI 1.2–5.6) but was decreased with methotrexate treatment (RR 0.4; 95% CI 0.2–0.9).
Table 2. Rate ratios of ILD associated with DMARD use in all subjects and in subjects stratified by previous methotrexate use and previous ILD during the year prior to entry into the cohort*
Previous interstitial lung disease (ILD) was either an outpatient diagnosis or a secondary inpatient diagnosisduring the year prior to cohort entry.
Subjects with no previous methotrexate use and no previous ILD†
Subjects with previous methotrexate use or previous ILD†
For the sensitivity analysis, the case definition that included secondary discharge diagnoses identified 246 cases. The resulting adjusted RRs for ILD associated with methotrexate treatment (1.0; 95% CI 0.7–1.3) and with leflunomide treatment (2.2; 95% CI 1.6–3.0) remained similar to those from the study case definition. For the 1,073 cases in which outpatient diagnoses were also included, the adjusted RRs for ILD associated with methotrexate treatment (1.1; 95% CI 0.9–1.2) and with leflunomide treatment (1.5; 95% CI 1.3–1.8) also remained comparable.
To address the issue of channeling, we identified all 64,676 subjects who started on a DMARD on or after September 1, 1998 and had at least 4 months of data available prior to the index DMARD prescription (Figure 1). Patients with a previous diagnosis of ILD, whether hospitalized or not, were almost twice as likely to have been prescribed leflunomide relative to methotrexate (adjusted OR 1.9; 95% CI 1.5–2.3) than patients with no such a diagnosis. These patients were also just as likely to have been prescribed biologic agents (OR 1.1; 95% CI 0.9–1.3) or traditional DMARDs (OR 1.0; 95% CI 0.8–1.3) compared with methotrexate.
In a large cohort of patients with RA treated with a DMARD, we observed an apparent 2-fold increased risk of ILD in patients treated with leflunomide compared with those not treated with leflunomide. However, the risk increase with leflunomide was restricted to patients with a history of methotrexate use or interstitial lung disease. There was no increase in the risk of ILD with leflunomide among the patients who had no previous methotrexate use and no interstitial lung disease prior to cohort entry.
The increased risk of ILD in the patients treated with leflunomide appears to be due to a channeling bias, where patients with a history of ILD may have been preferentially prescribed leflunomide rather than methotrexate on the assumption that, in contrast to methotrexate, no lung toxicity was associated with leflunomide. Indeed, in subgroup analyses, among patients defined by the absence of a history of methotrexate use or ILD, there was no increased risk of ILD with leflunomide treatment, while the risk with methotrexate treatment was increased 3-fold. On the other hand, among patients with a history of methotrexate use or ILD, methotrexate treatment actually appeared to be highly protective (RR 0.4; 95% CI 0.2–0.9) while the risk of ILD with leflunomide treatment was increased more than 2-fold. This channeling bias is further supported by the profile of patients who were prescribed leflunomide, where patients with a history of ILD were twice as likely to have been prescribed leflunomide as a first DMARD as compared with methotrexate, other traditional DMARDs, and biologic agents.
Identifying a possible association between leflunomide and lung toxicity is difficult in part because interstitial pneumonitis is so rare. In clinical trials of leflunomide and in subsequent observational studies, the incidence has been too low to assess the risk with precision, or the estimated risk was not increased compared with other DMARDs (1, 11–14). Large administrative databases have the advantage of allowing the identification of small risks that may have been missed in clinical trials.
The current study is limited by the fact that it was exclusively based on computerized claims data. It was thus impossible to validate the endpoint of ILD. However, we assumed that diagnoses recorded as primary diagnoses during the course of a hospitalization are more likely to be valid. A strength of the study is the inclusion of various populations in the cohort, encompassing Medicaid, Medicare, private health maintenance organizations, preferred provider organizations, and over 75 different managed care organizations, that make the findings generalizable.
We conclude that patients treated with leflunomide who have no history of methotrexate use and no existing ILD carry no excess risk of ILD. The increase in the risk of ILD associated with leflunomide is restricted to the subgroup of patients with a history of methotrexate use or existing ILD, and this appears to be the result of a channeling bias.
We thank Dr. Juhaeri Juhaeri of Sanofi-Aventis for assistance with the database acquisition.