1. Top of page
  2. Abstract
  9. Supporting Information


To determine whether the differences in the modes of action and safety profiles of individual tumor necrosis factor inhibitors (TNFi) translate into differential mortality risks, as investigated in etanercept, infliximab, and adalimumab.


Data on patients with rheumatoid arthritis (RA) identified in the Swedish Biologics Register (Anti-Rheumatic Therapy in Sweden [ARTIS]) in whom first-ever treatment with a biologic agent (etanercept [n = 2,686], infliximab [n = 2,027], or adalimumab [n = 1,609]) was initiated between 2003 and 2008 were linked to national Swedish registers to get information on deaths from any cause, demographic features, RA characteristics, comorbid conditions, and concurrent treatment at the start of TNFi treatment. Hazard ratios (HRs) were modeled using multivariable adjusted and weighted Cox models.


During 19,118 person-years of followup, 211 patients died (3.3%; 1.1 deaths per 100 person-years); 85% of the deaths occurred among patients who had been exposed to only one TNFi. We found no statistically significant difference in overall mortality rates across the exposure groups, regardless of adjustment and modeling approach (for infliximab versus etanercept, HR 1.1 [95% confidence interval (95% CI) 0.7–1.7], and for adalimumab versus etanercept, HR 1.3 [95% CI 0.9–2.0]).


Overall, we noted no statistically significant difference in mortality rates between the 3 TNF inhibitors under study. Further studies need to examine whether certain subsets of patients are at increased risk of death with specific TNFi.

Tumor necrosis factor α (TNFα) is a key mediator of the inflammatory response. Inhibition of TNF has been proven to be effective against rheumatoid arthritis (RA) and has been established as a mainstay alternative treatment of RA in clinical practice today. Treatment with a TNF inhibitor (TNFi) is prescribed in 15% or more of all patients with RA in Sweden (1). To date, 5 such treatments have been approved (etanercept in 1999, infliximab in 1999, adalimumab in 2003, golimumab in 2009, and certolizumab pegol in 2009; the latter two are not covered in this analysis).

Inhibition of TNF can be achieved through different routes. Apart from obvious differences, such as mode of administration (subcutaneous versus intravenous), etanercept, infliximab, and adalimumab are different with respect to their mode of action, half-life, and binding affinity. In particular, infliximab and adalimumab are monoclonal antibodies that can bind to extracellular, transmembrane, and receptor-bound TNFα. Etanercept, on the other hand, acts as a decoy receptor that binds to extracellular and transmembrane TNFα. The implications of these differences are not fully understood. In terms of efficacy, there seem to be no major differences between the 3 agents in RA, as observed in other chronic inflammatory conditions. In terms of drug retention, differences across the 3 TNFi have been reported in RA (2). In terms of safety, differences have been suggested in the safety profiles of the 3 TNFi for tuberculosis (3–6), herpes zoster (7), and for other opportunistic infections (7–9).

Whether these documented or purported differences between etanercept, infliximab, and adalimumab translate into differences in overall mortality rates following treatment of RA is unknown. Trial data are limited by small numbers, short followup times, and the absence of head-to-head comparisons. Observational studies have shown differences in mortality rates between TNFi-treated patients with RA and cohorts of patients treated otherwise, but the TNFi have not been compared to each other in this setting (10–12). Therefore, we investigated the association between specific TNFi and the risk of death from any cause. Specifically, using a large national population-based cohort of patients with RA initiating therapy with a TNFi as their first biologic disease-modifying antirheumatic drug (DMARD), we made comparisons of the risk of death among patients beginning treatment with etanercept, infliximab, or adalimumab.


  1. Top of page
  2. Abstract
  9. Supporting Information

Setting and data sources.

Medical care in Sweden is public and tax-funded, which in combination with the unique national registration number, facilitates linkage of data from different national registers and other data sources. Patients with RA predominantly seek care from rheumatologists, the vast majority of whom work within the hospital setting (13). Swedish rheumatology care, the Swedish healthcare setting, and register linkage have been described elsewhere in more detail (14, 15). For this study, we used the following data sources. The Swedish Biologics Register (Anti-Rheumatic Therapy in Sweden [ARTIS]) has collected data on adult patients prescribed biologic agents for the treatment of rheumatic diseases in Sweden since 1999. Patients included in this register because of a diagnosis of RA were identified, and the following data were collected: age, RA duration, date of initiating therapy with the first TNFi, disease activity, and any treatment discontinuation or switching (14). The coverage of the ARTIS database has been estimated to be nearly 90% of all eligible patients with RA (16).

The Prescribed Drug Register retains information on dates and drugs for all pharmacy dispensations nationwide from July 2005 onward. Drugs such as infliximab, which are administered by infusions in hospital settings, are not covered on a patient-level basis or are covered to a lesser extent; in 2009, 18% of all infliximab treatment in Sweden was provided through ambulatory care (17).

The Swedish National Patient Register contains data on hospitalizations (nationwide since 1987) and outpatient specialist visits since 2001. Discharge diagnoses are coded according to the calendar year–appropriate International Classification of Disease (ICD) codes. One primary discharge diagnosis and up to 7 contributory diagnoses can be reported at each visit. The completeness of the inpatient component of the register is >99% (18), while the coverage of the outpatient component varies with specialty (highest for somatic nonsurgical care). For primary discharge diagnoses at reported visits to specialist outpatient somatic care, the completeness was >87% (

The Cause of Death Register is virtually complete with information on >99% of all deaths in Sweden since 1952 (19).

The Register of the Total Population has information on sex, year of birth, residency, immigration, and emigration.

The Longitudinal Integration Database for Health Insurance and Labour Market Studies retrieves data on education from the Education Register. Data are collected annually on people under the age of 75 years.

The study population was identified as all patients with RA in the ARTIS database who initiated therapy with a first-ever TNFi (etanercept, infliximab, or adalimumab) from 2003 through 2008 (n = 6,322). Although data on biologic agents were available from the time of initial licensing in 1999, the present study focused on those in whom therapy was initiated in 2003 and onward, because 2003 was the year when all 3 TNFi of interest were first available. Furthermore, although the Prescribed Drug Register could be used to identify some additional initiators from mid-2005 onward, ARTIS followup data on disease activity, concomitant treatment, and other important factors for the present analyses would likely be missing for these patients.

The study was approved by the Ethics Committee at Karolinska Institutet.

Followup and outcome.

The outcome of interest was death from any cause, which was identified through linkage of the study population to the Swedish Cause of Death Register through December 31, 2008. For all individuals in the study population, followup time began at the first known use of etanercept, infliximab, or adalimumab. In the primary analysis, each individual contributed person-time until the first of the following possible events occurred: death, emigration, initiation of another biologic DMARD (TNFi or otherwise), or end of followup (December 31, 2008). Two alternative followup periods were considered. In the first alternative analysis, followup time ignored discontinuation or switching to another biologic DMARD. In the second alternative analysis, individuals could only contribute person-time up to 90 days after discontinuing their first TNFi. Through linkage to the Register of the Total Population, we identified individuals who emigrated before the end of the study period (n = 11). Their data were censored at the time of emigration.

Covariate data.

In addition to age and sex, measures of disease activity at the time of TNFi initiation were extracted from the ARTIS database, including the Disease Activity Score in 28 joints (DAS28) (20), C-reactive protein (CRP; mg/liter) level, erythrocyte sedimentation rate (ESR; mm/hour), tender and swollen joint counts, and disease duration. Disease duration was defined as a categorical variable (<5 years, 5–9 years, ≥10 years, missing). Scores on the patient's global assessment of overall health and pain (100-point scale) and functional status, as measured by the Health Assessment Questionnaire (HAQ) (21), were also available. Data on concurrent DMARD treatment were obtained from the ARTIS database and assessed in 2 different, but not mutually exclusive, ways: 1) treatment with any DMARD, including methotrexate, at the start of TNFi therapy and 2) treatment specifically with methotrexate at the start of TNFi therapy. Data on concomitant treatment with glucocorticoids were also collected from ARTIS.

Comorbid conditions were assessed through linkage of the study population to the Swedish National Patient Register. We identified hospitalizations or outpatient visits with primary or contributory discharge codes for the following conditions prior to initiation of TNFi therapy: diabetes mellitus (ICD-9 code 250; ICD-10 codes E10 and E11), chronic obstructive pulmonary disease (ICD-9 codes 491–493; ICD-10 codes J41–J46), acute myocardial infarction and unstable angina (ICD-9 codes 410 and 411B; ICD-10 codes I20 and I21), stroke and cerebrovascular disease (ICD-9 codes 430–438; ICD-10 codes I60–I69), and heart failure (ICD-9 code 428; ICD-10 code I50).

Previous joint surgery was identified using surgical procedure codes for knee, hip, foot, and shoulder procedures in the Swedish National Patient Register as well. This covariate served as an additional marker of RA severity.

Education, a proxy for socioeconomic position in Sweden, was defined as the highest education attained and was categorized as <10 years, 10–12 years, >12 years, or missing. Because this variable was collected among those younger than 75 years, missing data are correlated with older age.

Statistical analysis.

Etanercept, infliximab, and adalimumab treatment starts were defined as the 3 exposure categories of interest. The largest group (etanercept) was used as the reference category.

Participant characteristics at the start of TNFi therapy were summarized and compared across exposures. Differences were assessed by one-way analysis of variance, Kruskal-Wallis test, chi-square test, and Fisher's exact test, as appropriate. P values less than 0.05 were considered statistically significant. Missing values were incorporated using the missing indicator method. The missing indicators were retained in the analyses to guarantee that the study population was consistent across analyses. Missingness did not appear to be informative. As a second approach to evaluate the impact of missingness, we imputed median exposure-specific values for the missing values. Both approaches yielded nearly identical results.

Cox proportional hazards models were used to model the time until death among those initiating therapy with a TNFi. Three different definitions of exposure and risk windows were used. In the primary analysis, individuals were still considered at risk following discontinuation of their first TNFi but were artificially censored (i.e., no longer contributed person-time of followup) if (and when) they started a second biologic agent (a TNFi or a non-TNF drug). To account for the fact that artificial censoring at the time of switching may have introduced a selection bias, covariates believed to be predictors of treatment-switching and risk factors for mortality were adjusted for by using inverse probability of censoring weighting (IPCW), where weights were estimated as the inverse of an individual's probability of remaining exposed to only 1 TNFi (i.e., uncensored) (22). Robust variance estimates were used to calculate 95% confidence intervals (95% CIs) for the estimated hazard ratios in weighted models.

In the first alternative approach, exposure was defined as once exposed, always exposed, regardless of any subsequent treatment discontinuation or switches. In the second alternative approach, individuals were considered to be at risk from the start of their first TNFi until 90 days after any discontinuation of this drug (whether switching to a new biologic or not). In a sensitivity analysis, a 30-day lag period was also considered, and the results were similar.

All Cox models were stratified by county and adjusted for sex, age at first exposure, calendar year of TNFi initiation, RA disease duration, HAQ score, DAS28, ESR, education level, and concomitant methotrexate use at the start of followup. The assumption of proportional hazards was assessed by inspection of the Schoenfeld residuals, as well as by the Wald test of the interaction term of exposure and followup time (in both cases, the assumption appeared to be met; P = 0.59 for the latter approach in the main model). A number of stratified analyses were performed to assess the potential effect of modification of the exposure–mortality association with specific interest in comorbidity, concomitant therapy, and followup time. These models included stratification by sex, years of education completed (≤9 years, 10–12 years, >12 years, missing), presence or absence of comorbid conditions at the start of followup, concomitant treatment with a nonbiologic DMARD at the start of followup, calendar period of TNFi treatment start, and time since TNFi treatment start.

Sensitivity analyses excluding the first 30 days after the start of treatment were performed, as were sensitivity analyses (IPCW analysis) including lag periods of 30 and 90 days postswitch, during which events were attributed to the first TNFi agent. Analyses were performed using SAS 9.2 (SAS Institute) and Stata 11 (StataCorp) software.


  1. Top of page
  2. Abstract
  9. Supporting Information

A total of 2,686 patients started etanercept, 2,027 patients started infliximab, and 1,609 patients started adalimumab (Table 1). The year of treatment start differed across the 3 drugs, with an increasing proportion of patients starting etanercept over time. There were also modest yet statistically significant differences at baseline (treatment start) across the 3 groups. Infliximab initiators had shorter average disease durations but higher ESRs, while etanercept initiators had lower DAS28, HAQ scores, and CRP levels. Concomitant methotrexate use was more common among patients initiating treatment with infliximab. There were no significant differences in the presence of the comorbid conditions of interest by the time of initiation of the first TNFi between the 3 drug groups. However, there were significant differences in the history of joint replacement surgery prior to treatment exposure, with a smaller proportion of infliximab initiators having had prior surgery.

Table 1. Characteristics of 6,322 Swedish patients with RA at the initiation of a TNFi as the first biologic agent, 2003–2008*
 Etanercept (n = 2,686)Infliximab (n = 2,027)Adalimumab (n = 1,609)P
  • *

    RA = rheumatoid arthritis; TNFi = tumor necrosis factor inhibitor; HAQ = Health Assessment Questionnaire; DAS28 = Disease Activity Score in 28 joints; CRP = C-reactive protein; ESR = erythrocyte sedimentation rate; DMARDs = disease-modifying antirheumatic drugs; COPD = chronic obstructive pulmonary disease.

No. (%) female2,084 (78)1,519 (75)1,232 (77)0.1
Age at start, median years5758570.051
Education, no. (%)    
 ≤9 years619 (23)597 (29)467 (29)<0.0001
 10–12 years1,176 (44)909 (45)690 (43) 
 >12 years828 (31)458 (23)404 (25) 
 Missing63 (2)63 (3)48 (3) 
RA duration, no. (%)    
 0–4 years991 (37)855 (42)583 (36)0.0002
 5–9 years559 (21)391 (19)298 (19) 
 ≥10 years1,116 (41)761 (38)709 (44) 
 Missing20 (1)20 (1)19 (1) 
Tender joint count    
Swollen joint count    
HAQ score    
CRP, mg/liter    
ESR, mm/hour    
Pain score (100-point scale)    
Patient's global assessment    
Previous joint replacement surgery, no. (%)612 (23)403 (20)382 (24)0.01
 2003417 (38)473 (43)216 (20)<0.0001
 2004476 (40)342 (29)374 (31) 
 2005385 (39)301 (31)295 (30) 
 2006422 (40)355 (34)270 (26) 
 2007473 (47)294 (30)240 (24) 
 2008513 (52)262 (26)214 (22) 
Nonbiologic DMARDs, no. (%)    
 Any1,860 (69)1,776 (88)1,171 (73)<0.0001
 Combination DMARDs269 (10)220 (11)153 (10)0.4
 Methotrexate1,638 (61)1,645 (81)1,017 (63)<0.0001
 Steroids1,373 (51)882 (44)813 (51)<0.0001
Comorbid conditions, no. (%)    
 Diabetes mellitus145 (5.4)103 (5.1)71 (4.4)0.36
 COPD143 (5.3)99 (4.9)74 (4.6)0.55
 Heart failure48 (1.8)27 (1.3)28 (1.7)0.44
 Ischemic heart disease128 (4.8)76 (3.8)59 (3.7)0.12
 Cerebrovascular51 (1.9)51 (2.5)36 (2.2)0.35
 Cancer111 (4.1)88 (4.3)63 (3.9)0.81
Hospitalizations, median4440.62
Days in hospital, median1921190.80

During followup, 514 (19%) of the etanercept patients, 718 (35%) of the infliximab patients, and 408 (25%) of the adalimumab patients switched to another biologic DMARD (Figure 1). As expected, patients who were switched to another biologic DMARD during followup had higher levels of disease activity (DAS28, joint counts), worse functional status (HAQ scores), higher levels of inflammation, and reported more pain at the time they started their first biologic DMARD (see Supplementary Table 1, available on the Arthritis & Rheumatism web site at This was true for all 3 of the TNF inhibitors.

thumbnail image

Figure 1. Distribution of study participants, person-years (py) of treatment, and deaths (D) during followup in a cohort of 6,322 Swedish patients with rheumatoid arthritis initiating therapy with a tumor necrosis factor inhibitor (etanercept, infliximab, or adalimumab), stratified by drug and by whether patients switched to a second biologic disease-modifying antirheumatic drug (DMARD) during followup. The time attributed to the second exposure to a biologic DMARD is the difference between the end date of the first drug and the start date of the second drug. This does not take into account any additional drug starts or discontinuations.

Download figure to PowerPoint

Among the 6,322 patients in our entire study population and during a total of 19,118 person-years of followup (mean 3 years; median 3 years), 211 patients (3.3%) died. The large majority (n = 179; 85%) of deaths occurred among those exposed to only 1 TNFi (Figure 1). There was little difference by drug during the first 2 years, in terms of drug-specific deaths by time (illustration available upon request from the corresponding author). Neoplasms (ICD-10 codes C00–D48) were the most commonly reported underlying cause of death (n = 64 [30.3%]), followed by diseases of the circulatory system (codes I00–I99; n = 56 [26.5%]), the musculoskeletal system and connective tissues (codes M00–M99; n = 27 [12.8%]), and the respiratory system (codes J00–J99; n = 18 [8.5%]).

Primary analysis.

When artificially censoring patients who switched to a second biologic DMARD but otherwise disregarding discontinuations of the first TNFi, we observed 69 deaths during 6,482 person-years of etanercept-only exposure, yielding a crude incidence rate (IR) of 1.1 deaths per 100 person-years (95% CI 0.8–1.3). The crude IR among those exposed exclusively to infliximab was 1.2 deaths per 100 person-years (95% CI 0.9–1.5), and 1.4 deaths per 100 person-years (95% CI 1.0–1.8) for exposure to adalimumab. In the Cox models, the overall results (Table 2) and the stratified results (Figure 2) were largely, similar with a few exceptions (Figure 2).

Table 2. Multivariable adjusted HRs and 95% CIs under 3 alternative exposure time definitions for the first TNFi therapy and death*
 Etanercept initiators, no. of deathsInfliximab initiatorsAdalimumab initiators
No. of deathsHR (95% CI)No. of deathsHR (95% CI)
  • *

    Adjusted for age at initiation of tumor necrosis factor inhibitor (TNFi) therapy, sex, calendar year of TNFi initiation, duration of rheumatoid arthritis, Health Assessment Questionnaire score, Disease Activity Score in 28 joints, erythrocyte sedimentation rate, education level, concomitant methotrexate at the start of TNFi therapy, and history of ischemic heart disease, diabetes mellitus, heart failure, cerebrovascular disease, and chronic obstructive pulmonary disease at TNFi initiation (each considered separately), and stratified by county. Etanercept was used as the reference group. HRs = hazard ratios; 95% CIs = 95% confidence intervals.

  • Since the start of the first TNFi, disregarding discontinuations but censoring at the first switch.

  • Since the start of the first TNFi, disregarding discontinuations/switches.

  • §

    Since the start of the first TNFi, censoring at 90 days after the first discontinuation.

Primary69561.1 (0.7–1.7)541.3 (0.9–2.0)
Alternative 172771.2 (0.8–1.7)621.2 (0.9–1.7)
Alternative 2§29241.3 (0.7–2.3)251.4 (0.8–2.4)
thumbnail image

Figure 2. Relative risk of death and corresponding 95% confidence intervals, by initiation of the first tumor necrosis factor inhibitor (TNFi) (infliximab [INF] or adalimumab [ADA] versus etanercept [reference group]) among a cohort of Swedish patients with rheumatoid arthritis (RA) in a stratified analysis, censoring initiators at the time of the first switch (primary analysis). Results were adjusted for age at initiation of TNFi therapy, sex, calendar year of TNFi initiation, duration of RA, Health Assessment Questionnaire score, Disease Activity Score in 28 joints, erythrocyte sedimentation rate, education level, concomitant methotrexate at the start of TNFi, and history of ischemic heart disease, diabetes mellitus, heart failure, cerebrovascular disease, and chronic obstructive pulmonary disease at TNFi initiation (each considered separately), and stratified by county. Edu = education; aTNF = anti–tumor necrosis factor (monotherapy [mono] or combination therapy [combo]).

Download figure to PowerPoint

In sensitivity analyses, where the followup time for switchers was extended for 30 days and 90 days beyond the switch date and, hence, were attributed to the first treatment exposure, the results were comparable (data not shown).

Alternative analysis 1: exposure since the start of the first TNFi.

With this alternative approach (considering all person-time after the first TNFi was started, ignoring any discontinuations or switches), 72 deaths occurred during 7,615 person-years of followup (0.9 death per 100 person-years [95% CI 0.7–1.2]) among those initiating etanercept, 77 deaths occurred during 6,587 person-years (1.2 deaths per 100 person-years [95% CI 0.9–1.5]) among those initiating infliximab, and 62 deaths occurred during 4,916 person-years (1.3 deaths per 100 person-years [95% CI 1.0–1.6]) among those initiating adalimumab. Using etanercept as the reference group, we noted no significantly different rates of death among those initiating infliximab (HR 1.2 [95% CI 0.8–1.7]) or those initiating adalimumab (HR 1.2 [95% CI 0.9–1.7]) (Table 2).

Stratified analyses showed differences by calendar year of initiation, concomitant DMARD therapy, and history of comorbid conditions (see Supplementary Table 2, available on the Arthritis & Rheumatism web site at

Alternative analysis 2: exposure from the start of the first TNFi until 90 days after stopping.

In the “on drug” analysis with censoring at 90 days after stopping the first TNFi, no statistically significant differences were observed either overall or in the stratified analyses (Table 2 and Supplementary Table 2).

In all 3 modeling approaches, the results of sensitivity analyses starting followup at 30 days after initiation of the first TNFi were nearly identical.


  1. Top of page
  2. Abstract
  9. Supporting Information

In this nationwide and population-based study of mortality rates following treatment of RA with either etanercept, infliximab, or adalimumab in clinical practice, we found no statistically significant difference in overall mortality rates across the 3 agents examined. Mortality rates following treatment of RA with specific TNFi have previously been studied using data from randomized controlled trials (RCTs) (10, 11, 23, 24). Due to the small numbers of patients, individual RCTs of TNFi have typically not been able to evaluate death as an end point, nor have they been able to compare the different TNFi to each other.

In a pooled analysis of individual patient data examining TNFi safety and restricted to 18 other RCTs among 8,808 patients over 7,846 person-years of followup (mean followup time 307 days in the TNFi group and 285 days in the control arms), investigators found no statistically significantly increased risk of death associated with TNFi therapy (all 3 drugs combined) versus the control arms (relative risk 1.39 [95% CI 0.74–2.29]), based on a total of 35 deaths versus 11 deaths (23). The drug-specific relative risks were 2.34 (95% CI 0.67–8.12) for etanercept, 0.62 (95% CI 0.21–1.79) for infliximab, and 2.04 (95% CI 0.64–6.51) for adalimumab (23). The short followup and few events precluded distinguishing between drug-specific differences, true differences, and mere chance. Furthermore, it is not self-evident that mortality within the context of a clinical trial reflects the types and causes of death in clinical practice.

Previous observational studies have exclusively examined TNFi as a class, comparing mortality rates following treatment with a TNFi to the rates associated with other RA treatments, rather than comparing the individual TNFi to each other. In doing so, previous studies have addressed a slightly different research question. Although there were some differences in the study designs and characteristics of the patient populations across studies, the reported relative risks of death in these studies tended to be <1, although they were not always statistically significant (10–12, 24).

The main result of our study was that there was no overall statistically significant difference in mortality rates between the 3 TNF inhibitors examined. We found, however, several statistically significant differences between etanercept versus adalimumab and versus infliximab (separately) as we explored certain patient subsets. Given the large number of subset analyses performed (3 dichotomous variables and 3 3-category variables across 3 analyses, which yielded 90 statistical tests), we should expect chance as a potential explanation for at least 4 of the statistically significant stratum-specific risks (5 were observed). While we cannot exclude chance as playing a role in our findings, these subset differences are compatible with a number of additional interpretations. First, these were artificial differences related to the selection of patients to receive each of the 3 drugs, rather than to the effects of these drugs per se. Second, there was an interaction between individual TNFi and certain comorbid conditions on the risk of death. For example, if etanercept confers a higher risk of death in the presence (versus absence) of certain comorbid conditions, then the relative risks for infliximab and adalimumab (versus etanercept) might be more pronounced among patients with no such comorbid conditions. Third, the stratum-defining factors (year of start, comorbidity, and monotherapy) may be correlated and may therefore identify the same subgroup of patients. Last, the followup time was limited.

Our study has several strengths and limitations. Whereas the main threat to the validity of previous observational studies (10–12, 24) comparing class effects of TNFi therapy with other treatments is related to selection bias and confounding by indication (i.e., that patients selected to receive TNFi therapy may be qualitatively different from those selected to start or continue to receive other therapies) our study population consisted exclusively of individuals who were selected to start a first-ever TNFi. However, among patients selected for TNFi therapy, confounding by indication may still exist, for example, if patients selected to receive etanercept are different from those selected to receive infliximab or adalimumab. As shown in Table 1, there were differences in disease activity and disease characteristics between the 3 TNFi groups. We adjusted for many of these potential confounders in our analyses; however, we cannot exclude the possibility of bias due to residual or unmeasured confounding. “Comorbidity” was defined as hospitalizations listing any of 5 predefined conditions, but this may represent too blunt a measure to adequately reflect “frailty,” particularly when treated as an indicator in the stratified analyses. In the multivariable adjusted models, each comorbid condition was adjusted for separately. Additional adjustment for the number of days in the hospital or the number of visits and admissions did not appreciably alter our results. Furthermore, although all 3 agents were approved and available for use in RA patients in 2003, adalimumab was newly introduced into the market, while rheumatologists had had up to 4 years of experience with etanercept and infliximab at the start of the present study.

Whereas patients in clinical trials may be very different from those seen in clinical practice, our nationwide and population-based sample should be reflective of real-life patients. The censoring of individuals at the time of treatment switching in our primary analysis may have introduced some selection bias; however, the models were weighted to account for this. The data presented in Supplementary Table 1 (available on the Arthritis & Rheumatism web site at suggest that those who switched were indeed different from those who did not switch. These covariates were all accounted for in the generation of the weights, and many were also adjusted for in the Cox models.

Also, while we did have a large nationwide cohort of patients initiating TNFi therapy, the number of events varied between 78 and 211, depending on the type of analysis performed. Statistical power was limited, given the adjustment for numerous confounders and stratification. In sensitivity analyses, when the 21 Swedish counties were replaced by the 6 health care regions, the results were virtually unchanged. Furthermore, removing numerous potential confounding factors did not appreciably alter the estimated relative risks or corresponding interval estimates.

In conclusion, despite well-known differences in the mode of action as well as the documented differences in the safety profiles of the 3 TNF inhibitors examined (etanercept, infliximab, and adalimumab), these differences did not translate into statistically significant differences in overall mortality rates among 6,322 patients with RA treated in clinical practice. We cannot, however, exclude the possibility of differences in mortality rates across these 3 drugs in certain patient subsets.


  1. Top of page
  2. Abstract
  9. Supporting Information

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Simard had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Simard, Askling.

Acquisition of data. Neovius, Askling.

Analysis and interpretation of data. Simard, Askling.


  1. Top of page
  2. Abstract
  9. Supporting Information
  • 1
    Neovius M, Simard JF, Askling J. Nationwide prevalence of rheumatoid arthritis and penetration of disease-modifying drugs in Sweden. Ann Rheum Dis 2011; 70: 6249.
  • 2
    Hetland ML, Lindegaard HM, Hansen A, Podenphant J, Unkerskov J, Ringsdal VS, et al. Do changes in prescription practice in patients with rheumatoid arthritis treated with biological agents affect treatment response and adherence to therapy? Results from the nationwide Danish DANBIO Registry. Ann Rheum Dis 2008; 67: 10236.
  • 3
    Askling J, Fored CM, Brandt L, Baecklund E, Bertilsson L, Coster L, et al. Risk and case characteristics of tuberculosis in rheumatoid arthritis associated with tumor necrosis factor antagonists in Sweden. Arthritis Rheum 2005; 52: 198692.
  • 4
    Gomez-Reino JJ, Carmona L, Valverde VR, Mola EM, Montero MD, on behalf of the BIOBADASER Group. Treatment of rheumatoid arthritis with tumor necrosis factor inhibitors may predispose to significant increase in tuberculosis risk: a multicenter active-surveillance report. Arthritis Rheum 2003; 48: 21227.
  • 5
    Sichletidis L, Settas L, Spyratos D, Chloros D, Patakas D. Tuberculosis in patients receiving anti-TNF agents despite chemoprophylaxis. Int J Tuberc Lung Dis 2006; 10: 112732.
  • 6
    Wolfe F, Michaud K, Anderson J, Urbansky K. Tuberculosis infection in patients with rheumatoid arthritis and the effect of infliximab therapy. Arthritis Rheum 2004; 50: 3729.
  • 7
    Strangfeld A, Listing J, Herzer P, Liebhaber A, Rockwitz K, Richter C, et al. Risk of herpes zoster in patients with rheumatoid arthritis treated with anti-TNF-α agents. JAMA 2009; 301: 73744.
  • 8
    Mohan AK, Cote TR, Block JA, Manadan AM, Siegel JN, Braun MM. Tuberculosis following the use of etanercept, a tumor necrosis factor inhibitor. Clin Infect Dis 2004; 39: 2959.
  • 9
    Salmon-Ceron D, Tubach F, Lortholary O, Chosidow O, Bretagne S, Nicolas N, et al, for the RATIO Group. Drug-specific risk of non-tuberculosis opportunistic infections in patients receiving anti-TNF therapy reported to the 3-year prospective French RATIO registry. Ann Rheum Dis 2011; 70: 61623.
  • 10
    Jacobsson LT, Turesson C, Nilsson JA, Petersson IF, Lindqvist E, Saxne T, et al. Treatment with TNF blockers and mortality risk in patients with rheumatoid arthritis. Ann Rheum Dis 2007; 66: 6705.
  • 11
    Lunt M, Watson KD, Dixon WG, British Society for Rheumatology Biologics Register Control Centre Consortium, Symmons DP, Hyrich KL, on behalf of the British Society for Rheumatology Biologics Register. No evidence of association between anti–tumor necrosis factor treatment and mortality in patients with rheumatoid arthritis: results from the British Society for Rheumatology Biologics Register. Arthritis Rheum 2010; 62: 314553.
  • 12
    Carmona L, Descalzo MA, Perez-Pampin E, Ruiz-Montesinos D, Erra A, Cobo T, et al. All-cause and cause-specific mortality in rheumatoid arthritis are not greater than expected when treated with tumour necrosis factor antagonists. Ann Rheum Dis 2007; 66: 8805.
  • 13
    Svensk Reumatologisk Forening (SRF). Läkarbemanning för reumatologi mars 2009. ReumaBulletinen 2009; 73: 19.
  • 14
    Askling J, Fored CM, Geborek P, Jacobsson LT, van Vollenhoven R, Feltelius N, et al. Swedish registers to examine drug safety and clinical issues in RA. Ann Rheum Dis 2006; 65: 70712.
  • 15
    Askling J, van Vollenhoven RF, Granath F, Raaschou P, Fored CM, Baecklund E, et al. Cancer risk in patients with rheumatoid arthritis treated with anti–tumor necrosis factor α therapies: does the risk change with the time since start of treatment? Arthritis Rheum 2009; 60: 31809.
  • 16
    Neovius M, Simard J, Sundstrom A, Jacobsson L, Geborek P, Saxne T, et al for the ARTIS Study Group. Generalisability of clinical registers used for drug safety and comparative effectiveness research: coverage of the Swedish Biologics Register. Ann Rheum Dis 2011; 70: 5169.
  • 17
    Neovius M, Sundstrom A, Simard J, Wettermark B, Cars T, Feltelius N, et al. Small-area variations in sales of TNF inhibitors in Sweden between 2000 and 2009. Scand J Rheumatol 2011; 40: 815.
  • 18
    Socialstyrelsen. In-patient diseases in Sweden 1987-2003. Stockholm: National Board of Health and Welfare, Centre for Epidemiology; 2005.
  • 19
    Socialstyrelsen. Causes of death 2007. Stockholm: National Board of Health and Welfare, Centre for Epidemiology; 2009.
  • 20
    Prevoo ML, van 't Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity scores that include twenty-eight–joint counts: development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum 1995; 38: 448.
  • 21
    Fries JF, Spitz P, Kraines RG, Holman HR. Measurement of patient outcome in arthritis. Arthritis Rheum 1980; 23: 13745.
  • 22
    Hernan MA, Lanoy E, Costagliola D, Robins JM. Comparison of dynamic treatment regimes via inverse probability weighting. Basic Clin Pharmacol Toxicol 2006; 98: 23742.
  • 23
    Leombruno JP, Einarson TR, Keystone EC. The safety of anti-tumour necrosis factor treatments in rheumatoid arthritis: meta and exposure-adjusted pooled analyses of serious adverse events. Ann Rheum Dis 2009; 68: 113645.
  • 24
    Dixon WG, Hyrich KL, Watson KD, Lunt M, Symmons DP. Influence of anti-TNF therapy on mortality in patients with rheumatoid arthritis-associated interstitial lung disease: results from the British Society for Rheumatology Biologics Register [published erratum appears in Ann Rheum Dis 2011;70:1519]. Ann Rheum Dis 2010; 69: 108691.


  1. Top of page
  2. Abstract
  9. Supporting Information


Members of the ARTIS Study Group, in addition to the authors, are as follows: Drs. E. Baecklund, L. Cöster, H. Forsblad-d'Elia, N. Feltelius, P. Geborek, L. Jacobsson, L. Klareskog, S. Lindblad, S. Rantapää-Dahlqvist, T. Saxne, and R. van Vollenhoven.

The ARTIS Study Group conducts scientific analyses using data from ARTIS, the Swedish Biologics Register, which is run by the Swedish Society for Rheumatology. For the maintenance of this register, the Swedish Society for Rheumatology has received funding (independent of the conduct of the scientific analyses described herein) from Schering-Plough, Bristol-Myers Squibb, Wyeth, Abbott Laboratories, UCB, and Roche.

Supporting Information

  1. Top of page
  2. Abstract
  9. Supporting Information

Additional Supporting Information may be found in the online version of this article.

ART_34582_sm_SupplTables.doc181KSupplementary Table 1 and 2

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.