Risk and case characteristics of tuberculosis in rheumatoid arthritis associated with tumor necrosis factor antagonists in Sweden†
The authors were solely responsible for all data collection, analysis, and writing of the manuscript.
Because treatment with tumor necrosis factor (TNF) antagonists may increase the risk of tuberculosis (TB), and because knowledge of the risk of TB in rheumatoid arthritis (RA) not treated with biologics is scarce and of uncertain generalizability to low-risk populations, this study sought to determine the risk of TB among Swedish patients with RA.
Using data from Swedish nationwide and population-based registers and data from an ongoing monitoring program of TNF antagonists, the relative risks of TB in patients with RA (versus the general population) and of TB associated with TNF antagonists (versus RA patients not treated with biologics) were determined by comparing the incidence of hospitalization for TB in 3 RA cohorts and 2 general population cohorts from 1999 to 2001. We also reviewed the characteristics of all reported cases of TB in RA patients treated with TNF antagonists in Sweden and calculated the incidence of TB per type of TNF antagonist between 1999 and 2004.
During 1999–2001, RA patients who were not treated with TNF antagonists were at increased risk of TB versus the general population (relative risk 2.0, 95% confidence interval [95% CI] 1.2–3.4). RA patients treated with TNF antagonists had a 4-fold increased risk of TB (relative risk 4.0, 95% CI 1.3–12) versus RA patients not treated with TNF antagonists. The reported TB cases during 1999–2004 in RA patients exposed to TNF antagonists (9 infliximab, 4 etanercept, 2 both) were predominantly pulmonary. TB occurred up to 3 years following the start of treatment.
Irrespective of whether TNF antagonists are administered, Swedish patients with RA are at increased risk of TB. During 1999–2001, TNF antagonists were associated with an increased risk of TB, up to 4-fold in magnitude. This increased risk may persist over time during treatment and is related to both infliximab and etanercept.
Tumor necrosis factor (TNF) plays a major role in host defense against tuberculosis (TB) (1). The frequencies of TB in phase III trials of TNF antagonists (infliximab  and adalimumab ) have suggested that treatment with TNF antagonists (infliximab in particular) may indeed increase the risk of TB, and these observations have been followed by numerous spontaneous reports to regulatory agencies (4–6). Judgments on the significance of these descriptive data, however, require information on the incidence of TB in well-defined cohorts of patients with rheumatoid arthritis (RA) treated with TNF antagonists, in whom the risk of TB relative to the expected incidence of TB in contemporary RA patients not treated with biologics should be calculated.
Our current understanding of the relative risk of TB associated with RA per se and associated with TNF antagonists in routine care is scarce and can be derived from data on Spanish and US populations. In Spain, a 4-fold risk of TB in RA on the basis of 7 cases (7) and a further 12–20-fold increase in risk associated with TNF antagonists on the basis of 17 cases (8) were reported. In the US, a nonelevated risk of TB in RA in 1 case, but an increased incidence of TB in RA associated with TNF antagonists based on 4 cases (9), were reported. These cases of TB all occurred while the patients were receiving infliximab, and were characterized by a seemingly high proportion (50%) of extrapulmonary TB.
Of note, active TB in patients exposed to TNF antagonists mainly appears to constitute reactivation of latent TB (10), and is therefore a reflection of the pattern of infection many years previously. Accordingly, it is far from certain that the relative risks reported in Spain (8) or in ethnically heterogeneous populations in the US (9) are at all applicable to low-incidence areas such as Sweden or selected low-risk populations in the US. Indeed, the current incidence of TB in Spain (20 cases per 100,000 population) exceeds that in Sweden (5 cases per 100,000 population) and that in the non-Hispanic white population in the US (11). A proper judgment requires more data from populations with different background prevalences of TB. Likewise, although many case reports and case series (4, 6) have described the characteristics of TB occurring in TNF antagonist–treated patients (proportionally more occurring among those receiving infliximab compared with etanercept ), its origin in spontaneous reporting systems is obscured by underreporting and uncertainties regarding the underlying duration (person-time) of these drug regimens, thus limiting inferences and generalizability. To fully capture pretreatment risk factors for and characteristics of TB in TNF antagonist–treated individuals, data on consecutive cases of TB occurring in well-defined population-based cohorts of patients treated with TNF antagonists are needed.
In Sweden, high-quality nationwide health and census registers offer the unique potential to assess comorbidity rates through register linkage, which, at the time of this study, provided register data through 2001. Ongoing programs that monitor patients treated with TNF antagonists offer well-defined cohorts of such patients. We performed a cohort study to assess the relative risk of hospitalization for TB in patients with RA not treated with biologics, and the relative risk of hospitalization for TB in RA associated with TNF antagonists, from 1999 through 2001. In addition, we reviewed the medical files on all reported cases of TB among patients treated with TNF antagonists in Sweden during 1999–2004 and calculated drug-specific incidences of TB in that time period.
PATIENTS AND METHODS
We assembled 5 different cohorts for the study, as described below.
RA cohort from the Inpatient Register.
Swedish inpatient care is public and population-based, with referrals based on geography rather than financial capacity or health insurance. The Swedish Inpatient Register contains individual-based information on inpatient care for each county since 1964, with nationwide coverage since 1987. Discharge diagnoses (main diagnosis and up to 5 contributory diagnoses) for each discharge are coded according to the International Classification of Diseases (ICD) revisions 7–10 (13). We identified all individuals older than age 16 years who were discharged from inpatient care and identified as having RA at discharge between 1964 and 2001 (and not at any other point in time); patients discharged with a diagnosis of psoriatic arthritis, systemic lupus erythematosus, or ankylosing spondylitis were excluded. For each individual, we collected information on sex, date of birth, the national identification number (a 10-digit number unique to each Swedish resident ), and date of the first discharge listing RA (Table 1).
Table 1. Characteristics of the Swedish patients with rheumatoid arthritis (RA) followed up for tuberculosis (TB), 1999–2001
|No. male/no. female||8,149/23,036||724/1,706||665/1,835|
|Age at start of followup, no.|| || || |
| 0–39 years||5,949||384||460|
| 40–59 years||12,724||933||1,236|
| 60–79 years||11,195||1,011||763|
| 80+ years||1,317||102||41|
|Mean HAQ score*||–||0.8||1.5|
|Diabetes mellitus, %†||6.8||2.7||4.7|
|Hip/knee prosthesis, %†||23/16||3.9/1.7||23/12|
|Person-years of followup||77,431||7,600||4,000|
|TB cases, no.||27||2||4|
General population reference cohort from the Population Register.
For each of the individuals in the Swedish Inpatient Register RA cohort, we randomly selected 2 reference individuals from the Register of Total Population (the Swedish census register), who were matched to the RA patients by sex, year of birth, county of residence during the year of first discharge listing RA, and marital status (married, unmarried, widow). This reference cohort reflected the true background population.
General population reference cohort from the Inpatient Register.
The RA patients constituting the Inpatient Register RA cohort were identified on the basis of at least one overnight hospital stay listing RA between 1964 and 2001. To assess whether any increased risk in this cohort was due to the hospitalization rather than to the RA itself, we assembled a second reference cohort by randomly selecting, for every individual in the Inpatient Register RA cohort, 4 reference individuals from the entire Inpatient Register who had been hospitalized due to any medical diagnosis during the same year as that of the first discharge listing RA in the corresponding Inpatient Register RA patient. The reference subjects were further matched by sex, year of birth, and county of residence. This second reference cohort thus reflected the ever-hospitalized subset of the general population.
Early RA cohort.
A cohort of patients with early RA consisted of all individuals with newly diagnosed RA (<1 year) that had been diagnosed at either of the participating rheumatology outpatient clinics/units or departments in Sweden since the mid-1990s, with a geographically varying but increasing coverage of the estimated numbers of cases of incident RA (40–100% in different regions, overall ∼70%) (15). For each individual, date of birth, detailed medical information on the RA diagnosis, date of RA diagnosis, sex, and national identification number were collected (Table 1).
TNF antagonist RA cohort.
Within the context of an ongoing Swedish comprehensive postmarketing surveillance program (Anti-rheumatic Treatment in Sweden, or ARTIS), we assembled a cohort of 2,500 RA patients who were treated with etanercept (983 subjects, 1,722 person-years) or infliximab (1,565 subjects, 2,050 person-years) between 1999 and December 31, 2001 (Table 1). Details and patient identification methods have been described elsewhere (16). Briefly, patients were identified through the Swedish Medical Products Agency in collaboration with the Swedish Society for Rheumatology, and through regional surveillance programs of patients treated with TNF blockade. For each individual, we collected information on date of birth, sex, the national registration number, type of TNF antagonist, and start and discontinuation dates of treatment (Table 1).
During 1999–2001, i.e., the period during which we assessed relative risks of hospitalization for TB, the awareness of TB risks associated with TNF antagonists had received little clinical attention. (The first major report of TNF antagonist–associated TB was published in October 2001 ). Consequently, guidelines for pretreatment TB screening had not yet been set up, and although many patients might have had a pretreatment chest radiograph performed for various reasons, no routine and specific search for radiologic TB lesions was done, nor was there any routine recording of a history of TB. Performance of a purified protein derivative test was uncommon among these patients.
Followup and occurrence of TB.
Each individual in the 5 cohorts was linked to the Swedish Register of Total Population and Population Changes, 1969–2001, and to the Swedish Cause of Death Register, 1964–2001. Through this linkage, we collected information on the vital status of each individual from date of first discharge with RA until date of death/emigration or December 31, 2001. Through linkage of all individuals to the Swedish Inpatient Register, 1987–2001, we identified all discharges listing TB (ICD9 codes 010–019, ICD10 codes A15–A19). Each individual was counted only once, irrespective of the number of discharges listing TB.
To assess the relative risk of TB in patients with RA, we compared the incidence of hospitalization for TB in the Inpatient Register RA cohort with that in its 2 reference cohorts. First, to maximize statistical power, we compared incidences of hospitalization for TB during the followup period, 1987–2001. Thereafter, we restricted followup to 1999–2001 and included only those individuals in the Inpatient Register RA cohort who were hospitalized with RA from 1964 to 1997 and who did not have a hospitalization listing TB before January 1, 1999. This was done to mirror the period for which we had register-based followup data on the TNF antagonist cohort, and to minimize selection bias for inclusion in the Inpatient Register RA cohort due to hospitalization because of incipient TB. Irrespective of restriction, we kept the matched design intact. Values for relative risk were estimated using Cox regression.
To assess the relative risk of TB in RA associated with TNF antagonists, we compared the incidence of hospitalization for TB in the TNF antagonist cohort during 1999–2001 with that in the above-described Inpatient Register RA cohort followed up during 1999–2001, as well as with that in the Early RA cohort during 1999–2001. Individuals in the TNF antagonist cohort who appeared in either of the other 2 cohorts were censored from the latter cohorts on the date of entry into the TNF antagonist cohort. Relative risks in this cohort were estimated as age- and sex-adjusted incidence rate ratios.
Case review and incidence of TB following TNF antagonist treatment, 1999–2004.
In the second part of our study, we reviewed all reported cases of TB in RA patients treated with TNF antagonists in Sweden from 1999 through 2004. This review thus included the 4 cases involving hospitalization found in the register-based followup of 1999–2001 and all cases, irrespective of hospitalization, reported thereafter. Cases were reported to the Medical Products Agency pharmacovigilance database, to the national TNF antagonist surveillance database, or both. Scrutiny of the mandatory reports of TB cases to the Swedish Institute for Infectious Disease Control yielded 1 additional patient in whom TNF antagonist treatment was mentioned.
To estimate the incidence of reported TB during 1999–2004 among RA patients exposed to TNF antagonists, we divided the reported TB cases that occurred within the ARTIS program by the total and drug-specific number of person-years of followup in the same program.
Relative risk of hospitalization for TB in patients with RA versus general population, 1987–2001 and 1999–2001.
In analyses of the 1987–2001 followup period, the Inpatient Register RA cohort comprised 62,321 patients followed up for 467,770 person-years. During followup, 230 individuals in this cohort were hospitalized with TB, which, compared with the reference cohort from the Population Register, corresponded to a 4-fold increase in risk (relative risk 3.9, 95% confidence interval [95% CI] 3.1–5.0). When the Inpatient Register RA cohort was compared with the reference cohort drawn from the Inpatient Register, the risk of TB in RA was increased 60% (relative risk 1.6, 95% CI 1.3–1.9). These relative risks were essentially similar to those found in analyses restricted to 1999–2001 (Table 2), in which the Inpatient Register RA cohort encompassed 31,185 individuals (Table 1) and its reference cohort from the Inpatient Register encompassed 83,007 subjects, of whom 33 were hospitalized with TB.
Table 2. Relative risk of TB in RA cohorts compared with reference cohorts in Sweden, 1999–2001*
|Inpatient Register RA|| |
| Population Register reference||3.7 (1.7–8.1)|
| Inpatient Register reference||2.0 (1.2–3.4)|
|Anti-TNF–treated RA|| |
| Inpatient Register RA — no biologics||4.0 (1.3–12)†|
| Early arthritis — no biologics||4.1 (0.8–21)†|
Relative risk of hospitalization for TB in TNF antagonist–treated RA versus other RA, 1999–2001.
When the TNF antagonist–treated RA cohort was compared with either of the 2 cohorts of RA patients who were not treated with TNF antagonists, the 4 cases of hospitalization for TB in the TNF antagonist cohort corresponded to a 4-fold increased risk of TB (e.g., relative risk 4.0, 95% CI 1.3–12 versus Inpatient Register RA cohort not treated with biologics) (Table 2).
Case characteristics and incidence of TB associated with TNF antagonists, 1999–2004.
Between 1999 and September 2004, 15 cases of TB in RA patients treated with TNF antagonists were reported (Table 3). All but 1 case occurred during treatment with a TNF antagonist. Eleven patients had been treated with infliximab and 6 with etanercept (2 patients had received both; 1 of these 2 patients also received adalimumab). The mean duration of TNF antagonist treatment was 10 months (median 8 months). Ten patients were receiving concomitant treatment with methotrexate, and 8 with corticosteroids. Ten patients (67%) had pulmonary TB, 2 had disseminated (fatal) TB, and 3 had musculoskeletal TB. Review of the medical records did not reveal any known case of latent TB at the time of treatment initiation. None of the patients had received prophylactic treatment. Three patients originated from TB-endemic countries.
Table 3. Characteristics of TB occurring among anti-TNF–treated RA patients in Sweden between 1999 and September 2004*
Overall, the incidence of reported TB in RA treated with TNF antagonists during 1999–2004 was 118 (95% CI 58–210) per 100,000 person-years among RA patients during treatment, and 105 (95% CI 56–180) per 100,000 person-years among RA patients who had ever been treated with a TNF antagonist. Among RA patients who had received only infliximab the incidence (during treatment) of TB was 145 (95% CI 58–299) per 100,000 person-years; among RA patients who had received only etanercept the incidence was 80 (95% CI 16–232) per 100,000 person-years, and among patients who had received both infliximab and etanercept the incidence was 129 (95% CI 3.3–719) per 100,000 person-years. In comparing these incidences, the relative risk associated with etanercept was 0.5 (95% CI 0.1–2.4), which changed little after adjustment for age.
In our register-based assessment, patients with RA who were not treated with biologics were indeed at increased risk of TB compared with the general population, and TNF antagonist treatment was associated with a further 4-fold increased risk during 1999–2001. Characterization of all reported cases of TB occurring in RA patients treated with TNF antagonists in Sweden revealed that TB may occur not only shortly after the start of treatment, but also years thereafter. Moreover, the results suggested a higher incidence of TB associated with infliximab than with etanercept, although this was based on only a few cases.
The introduction of TNF antagonists has highlighted the need for appropriate data on comorbidity in treated conditions. Specifically, many adverse events that may limit the use of these agents may be difficult to detect and quantify using data from short-term clinical trials, which often encompass subjects and conditions substantially different from those encountered during provision of routine care. Pharmacovigilance based on spontaneous reporting is limited by underreporting, an unknown propensity among physicians to report adverse events and to select the cases reported, and uncertainties with respect to the sample size corresponding to the reported number of adverse events (17, 18). With respect to TB, it is unclear whether RA in conjunction with common treatments is, in itself, a risk factor. Indeed, it is likely that the incidence of TB in RA differs from that in otherwise-healthy individuals. In low-incidence populations, most cases of TB occur in individuals with particular risk factors (19). Therefore, national TB rates may not be suitable as comparator rates, especially not in ethnically heterogeneous populations (9, 11). Studies of TB are hindered by yet another methodologic complication, with implications for generalizability: since TNF-associated TB largely represents reactivation of latent infection (10), the burden of TB attributable to TNF antagonists may be a function of past prevalence of natural infection with TB in the birth cohorts that typically harbor patients with RA.
The relative risks of TB associated with RA in our study were more precise, but in the same order of magnitude as those reported from Spain (8). In contrast to the findings reported in the US cohort (9), our results suggest that RA patients are at increased risk of TB even in the absence of TNF antagonists. The further 4-folding of the risk among RA patients treated with TNF antagonists is lower than the 12–20-fold increased risk reported during the same period in Spain (8), and may be lower than the risk increase observed in the US (9). Both the Spanish and the US studies were, however, restricted to infliximab-treated RA patients.
Apart from the different populations covered, the methodologic differences between the present study, the Spanish study, and the US study are substantial. The Spanish study was based on a cohort of infliximab-treated RA patients, the coverage of which was not reported, and TB was assessed through physician's reporting of culture-positive TB. National TB rates and rates obtained in a TNF antagonist–naive RA cohort were used as reference. In the US study, RA patients were selected from the practices of some 900 rheumatologists, and were followed up through regular contacts rather than through register linkage. TB was identified on the basis of patient self-report, which was subsequently validated. US rates of TB were used as reference. In our study, the Inpatient Register RA cohort was identified through prospectively recorded data on discharge diagnoses. Validation of these RA diagnoses suggests that almost 90% of them reflect RA as defined by the American College of Rheumatology (formerly, the American Rheumatism Association) criteria (20), with the remainder chiefly being other rheumatic conditions (21).
We estimate that ∼50% of all patients with RA in Sweden in 2001 were included in our cohort according to our inclusion criteria. To circumvent biases related to hospitalization, one of our general population reference cohorts was also identified in the Inpatient Register, and thus these patients were also required to have at least one overnight hospital stay, irrespective of diagnosis. Based on sales statistics, we estimate the coverage of the TNF antagonist cohort to be ∼80% (higher for etanercept than infliximab). Accordingly, we consider it unlikely that the Inpatient Register RA cohort or the Early RA Register cohort would harbor unidentified cases of TNF antagonist–associated TB, especially since the relative risks of TB in RA during 1987–2001 and 1999–2001 were similar.
In the assessment of relative risk of TB during 1999–2001, we recorded hospitalizations for TB and had no way to validate the diagnoses made, except for the TB cases occurring in the TNF antagonist cohort and in the Early RA cohort, in whom all cases of hospitalized TB during 1999–2001 also reflected true TB. Although hospitalization is not mandatory for TB treatment, the general tendency toward hospitalization may be higher in Sweden than in the US. Moreover, hospitalization is common during the clinical examination (e.g., because of unexplained inflammation/malaise or to obtain findings on bronchoscopy), thus leading to TB diagnosis. Accordingly, 12 of the 15 patients (80%) reviewed for TB had been hospitalized because of their TB. The incidence of TB hospitalization in our general population reference cohort was 10 cases per 100,000 population, which is similar to the national Swedish notification rates of 4–14 cases per 100,000 population in the corresponding age groups (19), and the incidence of hospitalization for TB following TNF antagonist treatment in the register-based part of the study was of similar magnitude as the incidence of TB based on the reported and reviewed TB cases.
A comparison of the distribution of ICD codes (A15.0–A19.9) defining the TB cases in the general population and Inpatient Register RA cohorts revealed similar relative risks for each code (e.g., A15.0) and a typical (19) mix of 75% pulmonary TB and 25% extrapulmonary TB, which provides further evidence against systematic differences in the diagnostic processes and coding between the cohorts under study. Of note, we used the same definition and method of ascertainment of TB (which was independent of drug exposure and devoid of recall bias) in all cohorts. Therefore, even if our method of TB ascertainment in 1999–2001 may not be 100% sensitive and may not provide accurate estimates of incidence (which we therefore do not report), the corresponding relative risks should remain valid. In the event that the propensity for hospitalization were higher in the TNF antagonist cohort than in, e.g., the general population cohort, the relative risks would, however, be biased (upwards), but such bias does not preclude our conclusion of a lower relative risk of TB following TNF antagonist treatment than has been previously reported.
The incidence of reported TB in RA patients treated with TNF antagonists provides us with the opportunity to compare the incidence of TB following treatment with etanercept with that following infliximab within the same country. Although, based on a few cases, the higher incidence of TB following treatment with infliximab in our study supports previous observations of a higher number of reported cases of TB following infliximab treatment, our results nevertheless suggest that the difference in risk between the 2 drugs may not be as marked as previously thought. Our review of TNF antagonist–treated TB cases indicated several cases of extrapulmonary TB. Importantly, in addition to immediate TB cases, our series also included cases occurring years after the start of TNF antagonist treatment, which may indicate that the risk of reactivation of TB following TNF antagonist treatment is not transient.
We thank the ARTIS study group on biologics use in RA in Sweden, the Swedish RA Register, and all doctors and health professionals involved in reporting patients within the framework of these structured programs. We also thank Maud Rütting, Medical Products Agency, for help with the retrieval of medical files of the TB cases, and the National Board of Health and Welfare.