Does tumor necrosis factor α inhibition promote or prevent heart failure in patients with rheumatoid arthritis?




To determine the hazard risk of developing or worsening heart failure in rheumatoid arthritis (RA) patients treated with tumor necrosis factor α (TNFα) inhibitors.


RA patients ages 18–75 years who started treatment with infliximab, etanercept, or adalimumab (n = 2,757), or conventional disease-modifying antirheumatic drugs (controls; n = 1,491) at the time of enrollment in a German biologics register were studied. Cox proportional hazards models were applied to investigate the influence of disease-related and treatment-specific risk factors on the incidence or worsening of heart failure.


The 3-year incidence rates of heart failure in patients with and patients without cardiovascular disease at the start of treatment were 2.2% and 0.4%, respectively. After adjustment for traditional cardiovascular risk factors, an increased risk of developing heart failure was found in patients who had a higher 28-joint Disease Activity Score at followup (hazard ratio [HR] 1.47 [95% confidence interval 1.07–2.02], P = 0.019). A residual nonsignificant risk related to treatment with TNFα inhibitors remained (adjusted HR 1.66 [95% confidence interval 0.67–4.1], P = 0.28). This residual risk was balanced by the efficacy of the anti-TNF treatment. When only baseline characteristics were taken into account, the HR related to TNFα inhibitor treatment decreased to 0.70 (95% confidence interval 0.27–1.84).


The findings of this study indicate that TNFα inhibitor treatment that effectively reduces the inflammatory activity of RA is more likely to be beneficial than harmful with regard to the risk of heart failure, especially if there is no concomitant therapy with glucocorticoids or cyclooxygenase 2 inhibitors. Furthermore, the data suggest that TNFα inhibition does not increase the risk of worsening of prevalent heart failure.

Rheumatoid arthritis (RA) is characterized by persistent inflammatory synovitis, joint destruction, and loss of functional capacity. RA is associated with an increased risk of cardiovascular morbidity and mortality (1–3); systemic inflammation is assumed to play an important role in the increased risk of atherosclerosis, myocardial infarction, heart failure, or cerebrovascular disease (1, 4–6). Inflammatory cytokines, including tumor necrosis factor α (TNFα), contribute to the pathogenesis of RA. TNFα is also a mediator of endothelial dysfunction, vascular instability, and disease progression in atherosclerosis (7, 8). Furthermore, it is known to contribute to the progression of heart failure (9, 10). Inhibition of TNFα has opened important new treatment options in RA. Treatment with infliximab, etanercept, or adalimumab has been shown to be effective in reducing signs and symptoms of the disease and in preventing joint damage (11–16).

However, etanercept or infliximab treatment for severe heart failure has not been successful (17, 18). Three trials of these TNFα inhibitors had to be halted prematurely, because neither etanercept nor infliximab was superior to placebo with respect to the composite score (mortality or hospitalization because of heart failure) (17, 18). An even higher heart failure risk than that observed with placebo was found in the high-dose infliximab group (10 mg/kg body weight) (18). This raised concerns about the heart failure risk in RA patients treated with infliximab, etanercept, or adalimumab.

In the study described herein, we used data from the German biologics register RABBIT (German acronym for Rheumatoid Arthritis—Observation of Biologic Therapy), an ongoing, closely monitored, prospective cohort study (19), to investigate the risk factors for heart failure in RA, and in particular the possible contribution of anti-TNF treatment. We hypothesized that, in addition to known risk factors in the general population, the severity and inflammatory activity of the rheumatic disease have an influence on the development or worsening of heart failure, as does treatment with TNFα inhibitors, glucocorticoids, and possibly selective cyclooxygenase 2 (COX-2) inhibitors.


Data source.

The ongoing German prospective cohort study RABBIT was established in May 2001 to investigate the long-term safety and efficacy of etanercept, infliximab, adalimumab, and anakinra in comparison with conventional disease-modifying antirheumatic drugs (DMARDs) in daily rheumatologic care (19). Patients ages 18–75 years who meet the American College of Rheumatology (formerly, the American Rheumatism Association) criteria for RA (20) are eligible as cases if new treatment with infliximab, etanercept, anakinra or adalimumab is started or as controls if a conventional DMARD therapy is started after failure of at least 1 other DMARD (see refs. 19 and21–23 for further details). The study protocol was approved by the Ethics Committee of Charité University School of Medicine.


RA patients enrolled in the RABBIT study between May 1, 2001, and May 1, 2006, who were treated with adalimumab, etanercept, infliximab, or conventional DMARDs, were included in the present study. Patients receiving anakinra (n = 83) were excluded because of the small size of this subgroup and the different mode of action of this agent. Patients who switched to anakinra or rituximab were considered at risk only for the time preceding this treatment switch. Possible study dropout status was checked for all patients who missed 2 or more consecutive followup visits, by contacting the treating physicians and, if necessary, the patients themselves. Reasons for dropout were ascertained.


At baseline and at 3-, 6-, 12-, 18-, 24-, 30-, 36-, 48-, and 60-month followup, the number of tender and swollen joints (28-joint count) (24), erythrocyte sedimentation rate, C-reactive protein (CRP) level, and duration of morning stiffness were recorded at the treating rheumatologist's office. The physician also recorded data on the start/end of DMARD and/or biologic therapy, reasons for treatment termination, concomitant therapy with glucocorticoids, nonsteroidal antiinflammatory drugs, or COX-2 inhibitors, and serious and nonserious adverse events. For consistency and quality assurance, the rheumatologists were provided with definitions of serious and nonserious adverse events according to the International Conference on Harmonization E2A guideline (25). They were asked to report every new symptom in the patients regardless of whether it was considered to be related to the treatment or not, as well as every instance of worsening of symptoms of a preexisting comorbid condition. One of the authors (AS) coded the adverse events using the Medical Dictionary for Regulatory Affairs, version 9.0 ( Disability was assessed with the Hannover Functional Status Questionnaire (Funktionsfragebogen Hannover [FFbH]) (26, 27), and the Disease Activity Score 28-joint assessment (DAS28) was calculated (28). The following comorbidities were included in the risk assessment (from a list of 25 comorbid conditions assessed by the treating physician at study entry): heart failure, coronary heart disease, cerebrovascular disease, hypertension, hyperlipoproteinemia, type 1 diabetes mellitus, type 2 diabetes mellitus, chronic obstructive pulmonary disease, and any other chronic lung disease. In order to achieve robust results, these comorbid conditions were combined into 4 groups: heart failure, any other cardiovascular disease, diabetes, and chronic lung disease.

Outcome measures.

The objective of this analysis was to estimate the incidence of and risk factors for heart failure in RA. A statistical analysis plan was specified in advance, in February 2006, after discussion with the scientific advisory board of the RABBIT study. According to this plan, all adverse events reported as heart failure, acute heart failure, congestive heart failure, or ventricular failure between May 1, 2001 and December 1, 2006 were counted. Symptoms alone, such as exertional dyspnea or peripheral edema, were not counted: in order for heart failure to be recorded, there had to be a formal diagnosis. All participating rheumatologists are specialists in internal medicine and are familiar with heart failure diagnosis. However, since no uniform criteria for the diagnosis of heart failure were included in the case report forms, we went back to the treating physician whenever heart failure was recorded as an adverse event. The physician was asked to confirm the diagnosis and to report further details of the diagnostic procedures, fulfillment of the New York Heart Association (NYHA) functional classification criteria (29), and concurrent cardiovascular diseases. Similarly, reports of worsening of the severity of heart failure had to be confirmed.

Statistical analysis.

Cox proportional hazards models with time-dependent covariates were applied to identify risk factors for heart failure and to address time-dependent parameters (activity of the disease, treatment). Hazard ratios (HRs) and 95% confidence intervals (95% CIs) were calculated. In the first step, we identified known risk factors in the general population that were traceable in our data and therefore could be used for adjustment. We considered age, sex, body mass index, and the 4 groups of comorbid conditions noted above. All possibly associated risk factors (P < 0.15 by Wald's test) were used for adjustment in the subsequent steps. We did not include smoking as a risk factor in the models, because the data were available only for the subsample of patients who had already passed the month-24 visit, and in this subsample the smoking habits among patients receiving anti-TNF agents were similar to those found in the control group (see below).

In the second step we investigated the influence of disease-related and treatment-specific risk factors on the incidence/worsening of heart failure. The following baseline characteristics were included in the risk assessment: disease duration, rheumatoid factor status, and (as measures of disease severity) functional capacity (as measured with the FFbH), rheumatoid nodules, erosive disease (yes/no), and the propensity score. The propensity score (likelihood of being treated with biologic agents) was estimated by means of logistic regression with the covariates: age, sex, DAS28, FFbH, rheumatoid nodules, and erosive disease. The propensity score model fitted quite well. No significant differences between observed and predicted deciles at risk were observed (P = 0.8 by Hosmer-Lemeshow test). The propensity score model also had good discriminative power. Calculation of the corresponding c-statistic resulted in a value of 0.73.

Measurements obtained at followup were used to determine the influence of treatment and RA disease activity. Instead of using point estimates, a risk-window approach was considered more suitable for addressing the time changing effects of the activity and the treatment of the rheumatic disorder. Therefore, at each point in time, the mean of the DAS28 score, the mean of the CRP values, or the mean of the prednisolone dosages from the previous 2 visits were considered as risk indicators of the activity of the disease or the influence of treatment with glucocorticoids. Treatment with COX-2 inhibitors at these visits was coded as 0 (never), 0.5 (partly [at 1 visit]), or 1 (continuously).

Since we had complete data on the start and stop dates of every treatment with biologic agents, we were able to assess the influence of anti-TNF therapy more exactly. We counted a patient as being at risk of anti-TNF treatment at a specified point in time if she or he had received the treatment within the 6 months before this time point. In a sensitivity analysis, we additionally considered a 3-month risk window of anti-TNF treatment. RA-specific factors possibly associated with the outcome measure (P < 0.15 by Wald's test) were included in the final stepwise multivariate Cox regression analysis. To estimate the influence of anti-TNF treatment, this parameter was included in all models investigated. In an ad hoc analysis, we additionally considered a simplified Cox regression model. The objective was to estimate an overall risk of anti-TNF treatment as a combined risk comprising a possible beneficial effect (via the efficacy of these agents) and a possible harmful effect (direct risk of the agents). In this model we therefore excluded the time-varying covariates but included the propensity score to adjust for differences at baseline. The proportion of followup time during which anti-TNF agents were received (on a 0–1 scale) was used to estimate the overall effect of this treatment. The incidence or worsening of heart failure was estimated by the Kaplan-Meier method.

A nested case–control study was performed to compare the followup data on patients (incident cases) who developed heart failure during followup with data on matched controls. Cases and controls did not have prevalent heart failure at study entry. They were matched with regard to sex and history of any other cardiovascular disease at baseline (yes/no), and age, body mass index, and followup time had to be similar. The patient who was most similar to a case in terms of Mahalanobis metric was selected as the matched control. Wilcoxon's test and McNemar's test were used to compare cases with matched controls. The Mann-Whitney U test and chi-square test were applied to compare the baseline characteristics of the anti-TNF group and the control group.


Baseline characteristics of the patients.

A total of 4,248 patients fulfilled the inclusion criteria. Of these patients, 2,757 had started treatment with TNFα inhibitors (anti-TNF group) and 1,491 had started a new DMARD therapy at baseline (control group) (Table 1). Overall, the patients in the anti-TNF group were younger, had significantly more active disease, and were more limited in activities of daily living compared with the control group. The mean FFbH score, reflecting the percent of full function, was 56.5 in the anti-TNF group and 66.2 in the control group. Minor differences were observed with regard to comorbidity status and smoking habits. The proportion of patients with prevalent heart failure was higher in the anti-TNF group. Eight of 98 patients with heart failure were categorized in functional class III according to the criteria of the NYHA. Seven of these 8 patients were treated with TNFα inhibitors. No patient fulfilled the criteria for NYHA functional class IV.

Table 1. Baseline characteristics of the patients*
 Anti-TNF–treated (n = 2,757)Control (n = 1,491)P
  • *

    Except where indicated otherwise, values are the number (%). For some parameters, there were small numbers of missing values. Anti-TNF = anti–tumor necrosis factor; IQR = interquartile range; ESR = erythrocyte sedimentation rate; CRP = C-reactive protein; DAS28 = Disease Activity Score 28-joint assessment; FFbH = Hannover Functional Status Questionnaire (measuring functional capacity as a percentage of full function); DMARDs = disease-modifying antirheumatic drugs; COX-2 = cyclooxygenase 2.

Age, mean ± SD years53.7 ± 12.456.1 ± 11.5<0.00001
Female2,152 (78.1)1,177 (78.9)0.504
Disease duration, median (IQR) years9 (5–16)6 (3–12)<0.00001
Rheumatoid factor positive2,219 (80.5)1,069 (71.7)<0.00001
Swollen joint count (range 0–28), mean ± SD9.3 ± 6.16.8 ± 5.4<0.00001
ESR, median (IQR) mm/hour30 (16–48)22 (12–40)<0.00001
CRP, median (IQR) mg/liter18 (8–40)13 (6–27)<0.00001
DAS28, mean ± SD5.8 ± 1.25.1 ± 1.3<0.00001
FFbH, mean ± SD56.5 ± 22.866.2 ± 21.6<0.00001
No. of previous DMARDs, mean ± SD3.6 ± 1.41.9 ± 1.1<0.00001
COX-2 inhibitor treatment695 (25.6)323 (22.0)0.010
Glucocorticoid treatment2,302 (83.9)1,132 (76.1)<0.00001
 Prednisolone ≥10 mg/day902 (32.9)297 (20.0)<0.00001
 Heart failure75 (2.7)23 (1.5)0.014
 Coronary heart disease149 (5.4)105 (7.0)0.033
 Cardiovascular disease total1,026 (37.3)569 (38.2)0.566
 Diabetes226 (8.2)128 (8.6)0.673
 Chronic lung disease201 (7.3)95 (6.4)0.256
Smoking (n = 1,973)   
 Current291 (23.3)167 (23.0)0.886
 Never668 (53.5)385 (53.1)0.856

At followup, 101 patients (2.4%) had died, 14 due to heart failure. Four hundred twelve patients (9.7%) dropped out, and 168 (4.0%) had not attended the last 2 or 3 followup visits. The annual loss-to-followup rate (with loss-to-followup defined as dropping out or failure to attend the last 2 or 3 visits) was 5.1%. The annual dropout rate was 3.9% on average, and the total dropout rate at 48 months was 15.5% (Kaplan-Meier estimate). Patients who dropped out did not differ significantly from those who completed the study with regard to age, sex, treatment with TNFα inhibitors, or cardiovascular disease status, but they did have slightly more active disease at the start of treatment (mean ± SD DAS28 score 5.7 ± 1.3 versus 5.5 ± 1.3).

Incidence of heart failure and rates of worsening of prevalent heart failure.

Twenty-five patients developed heart failure for the first time during the study period (Figure 1A). The incidence rates differed significantly between patients with cardiovascular disease at baseline (3-year incidence 2.2% [95% CI 1.3–3.5%]) and patients without cardiovascular disease at baseline (3-year incidence 0.4%). In 7 of the 25 patients with incident heart failure, the heart failure resulted in death (Figure 1B). A worsening in severity of prevalent heart failure was observed in 12 of 98 patients (3-year incidence 12.5%). This resulted in death in a further 7 patients, corresponding to a 3-year mortality rate of 9.0% (Figure 1B).

Figure 1.

Rates of incident heart failure and worsening in the severity of prevalent heart failure (A), and mortality due to heart failure (B). Oth. = other; CVD = cardiovascular disease (see Patients and Methods). Values in parentheses are the 95% confidence intervals of Kaplan-Meier estimates at 36 months.

Comparison of heart failure cases with matched controls.

In the analysis comparing patients who developed heart failure de novo (cases) with the remaining patients in the total sample and with matched controls, it was found that the group of cases was on average 9 years older than the remaining patients and had a significantly higher percentage of males, significantly greater weight, and highly significantly lower functional capacity and greater level of disease activity than patients without heart failure (Table 2). Matched controls, who were similar to the heart failure cases with respect to non–RA-specific risk factors (age, sex, comorbid cardiovascular conditions, body mass index) and followup time, had significantly lower functional capacity than the overwhelming majority of the other patients, probably because of their older age and greater frequency of cardiovascular comorbidities. More importantly, these controls had a clinically relevant and statistically significantly lower mean DAS28 score at followup (4.4) compared with the incident heart failure cases (5.1) (Table 2).

Table 2. Characteristics of the patients who had heart failure at followup (cases) compared with the total sample of remaining patients and with matched controls*
 Heart failure status at baseline
No heart failurePrevalent heart failure
Total remaining patients (n = 4,125)Incident heart failure cases (n = 25)Matched controls (n = 25)PTotal remaining patients (n = 86)Patients whose heart failure worsened (n = 12)
  • *

    Except where indicated otherwise, values are the number (%). For some parameters, there were small numbers of missing values. See Table 1 for definitions.

  • Incident cases versus matched controls.

  • For incident cases, last visit before heart failure; for matched controls, same visit as the corresponding case; for the total sample of patients without incident heart failure, mean at 12 months.

 Age, mean ± SD years54.2 ± 12.163.0 ± 9.062.2 ± 9.80.6166.9 ± 6.967.3 ± 5.3
 Male878 (21.3)8 (32.0)8 (32.0)24 (27.9)9 (75)
 Any other cardiovascular disease1,477 (35.8)19 (76.0)19 (76.0)  
 Body mass index, mean ± SD kg/m226.1 ± 5.128.9 ± 7.829.0 ± 6.328.0 ± 5.627.6 ± 5.0
 FFbH, mean ± SD60.5 ± 22.635.8 ± 20.546.2 ± 19.90.0540.9 ± 22.350.5 ± 26.2
 DAS28, mean ± SD5.5 ± 1.36.4 ± 1.25.9 ± ± 1.25.7 ± 1.0
 CRP, median (IQR)16 (7–35)23 (10–60)29 (16–48)0.8521 (8–51)33 (10–69)
 Anti-TNF treatment2,666 (64.6)16 (64.0)20 (80.0)0.3466 (76.7)9 (75.0)
 Glucocorticoids ≥10 mg/day1,155 (28.0)8 (32.0)9 (36.0)1.030 (34.9)6 (50.0)
 COX-2 inhibitor treatment983 (23.8)11 (44.0)3 (12.0)0.0422 (26.2)2 (16.7)
 DAS28, mean ± SD4.0 ± 1.55.1 ± 1.44.4 ± ± 1.44.5 ± 1.4
 CRP, median (IQR)8 (4–17)17 (8–33)12 (5–36)0.8814 (8–27)14 (9–45)
 Glucocorticoids ≥10 mg/day398 (11.6)5 (20.0)2 (8.0)0.4513 (18.1)6 (50.0)
 COX-2 inhibitor treatment592 (14.4)7 (28.0)4 (16.0)0.3812 (16.7)1 (8.3)
 Anti-TNF treatment1,877 (57.2)17 (68.0)20 (80.0)0.5143 (62.3))9 (75.0)

Patients with prevalent heart failure at the time of enrollment were on average 67 years old and had very low functional capacity and highly active disease at the start of treatment. At study entry they had a significantly higher frequency of TNFα inhibitor treatment (P = 0.01) than patients without this disorder. In 12 of these patients, the heart failure worsened during the followup period. This latter group had a significantly greater percentage of males and a significantly greater proportion who were receiving high-dose glucocorticoids compared with the remaining patients with prevalent heart failure.

Risk of heart failure developing de novo or worsening in severity.

Applying multivariate Cox proportional hazards regression analysis to the non–RA-specific risk factors, we found significant associations of heart failure with comorbid cardiovascular conditions, age, and sex, and possible associations with body mass index and chronic lung disease (Table 3), whereas no significant association was found for diabetes (P = 0.55). Therefore, age, sex, heart failure, presence of any other cardiovascular disease at baseline, chronic lung disease, and body mass index were used for multivariate adjustment of disease-related and treatment-specific risk factors in the second step of the analysis.

Table 3. Adjusted hazard ratios for heart failure*
 Adjusted for age, sex, heart failure, CVD, body mass index, and chronic lung diseaseMultivariate analysis final modelSimplified model to offset the efficacy of anti-TNF treatment
Adjusted HR95% CIPAdjusted HR95% CIPAdjusted HR95% CIP
  • *

    HR = hazard ratio; 95% CI = 95% confidence interval (see Table 1 for other definitions).

  • By Wald's test.

  • Referent was patients without cardiovascular disease (CVD) at study entry.

Characteristics at study entry         
 Age (per 10 years)1.641.08–2.500.0211.661.08–2.550.0211.741.10–2.740.018
 Male sex2.841.46–5.550.00224.042.03–8.02<0.00013.771.88–7.570.0002
 Patients with heart failure23.888.04–70.90<0.000115.995.40–47.36<0.000118.066.10–53.49<0.0001
 Patients with any other CVD3.471.33–9.030.0113.141.21–8.130.0183.091.19–8.040.021
 Body mass index (per 5 units)1.350.99–1.840.0611.421.05–1.910.0231.401.02–1.920.037
 Chronic lung disease1.870.84–4.160.13      
 Disease duration (per 5 years)1.090.94–1.270.28      
 Rheumatoid factor positive2.400.73–7.870.15      
 Rheumatoid nodules1.680.84–3.340.14      
 Erosive disease2.280.69–7.570.18      
 DAS28 (per 1 unit)1.270.96–1.680.091      
 FFbH (per 10-unit improvement)0.780.67–0.900.00080.820.70–0.960.0130.820.68–0.990.041
 Propensity score (per 0.1 unit)1.411.11–1.810.0057   1.200.88–1.640.25
Characteristics at followup         
 DAS28 (per 1 unit)1.501.15–1.950.00261.351.02–1.780.033   
  15–<30 mg/liter vs. <15 mg/liter1.280.52–3.130.60      
  ≥30 mg/liter vs. <15 mg/liter2.080.83–5.220.12      
  5–9 mg/day vs. <5 mg/day1.260.58–2.730.56      
  ≥10 mg/day vs. <5 mg/day2.050.82–5.090.12      
 COX-2 inhibitors2.150.99–4.690.054      
Anti-TNF vs. conventional DMARDs1.850.88–3.900.111.490.70–3.180.310.650.29–1.440.29

When considering separately the contributions of the various disease activity, severity, and treatment variables to heart failure risk, highly significant associations were found for functional capacity at the start of treatment, propensity score (likelihood of being treated with biologic agents), and DAS28 at followup (Table 3). The mean of the last 2 DAS28 scores before the onset of heart failure was more strongly associated with heart failure risk (HR 1.50 [95% CI 1.15–1.95], P = 0.0026) than was the DAS28 score before the start of treatment. Patients with a higher propensity score had a significantly higher risk of developing heart failure. A possible increase in the heart failure risk was also found for individual RA severity markers: rheumatoid factor, rheumatoid nodules, and erosive disease. Nonlinear associations were observed for CRP level and prednisolone dosage (Table 3).

Stepwise multivariate Cox regression analysis was performed to investigate the confounding of results for disease activity parameters by immunosuppressive therapy, and vice versa. Functional capacity and disease activity (DAS28) at followup remained significant predictors of heart failure (Table 3). An increase of 1 unit or 2 units in the DAS28 score resulted in a 1.4-fold or 1.8-fold increase, respectively, in the risk of developing heart failure. A 10% improvement in functional capacity was related to an 18% reduction in heart failure risk. The propensity score was clearly correlated with functional capacity (Pearson's r = −0.50) and partly with the DAS28 at followup (r > 0.2); therefore, the propensity score HR did not reach statistical significance in the stepwise Cox regression analysis. After adjustment for DAS28 score, function, and traditional risk factors, the HRs of treatment with high-dose (≥10 mg) and medium-dose versus low-dose (<5 mg) glucocorticoids decreased to 1.46 (95% CI 0.57–3.76) and 1.09, respectively, and did not achieve statistical significance. On the other hand, inclusion of glucocorticoid treatment and COX-2 inhibitor treatment (HR 1.99 [95% CI 0.90–4.37]) in the final model as possible confounders would have changed the HR for the DAS28 only slightly, to 1.31. No significant associations were found for anti-TNF therapy (Table 3).The length of the risk window for anti-TNF treatment had only a minor influence on the results (adjusted HR for 3-month risk window in final model 1.42 [P = 0.35]).

To estimate the effect of the treatment decision rather than the direct risks of the drugs, we additionally applied a simplified Cox regression model (see Patients and Methods). In this model we excluded the DAS28 at followup to offset the positive and negative effects of treatment. We found an overall protective effect of a decision to treat patients continuously with anti-TNF agents, but this was not statistically significant (Table 3).

Risk of heart failure developing de novo.

After exclusion of the prevalent heart failure cases, adjusted HRs for de novo development of heart failure were calculated (Table 4). These HRs were similar to those found in the previous analysis (Table 3), in which worsening of heart failure was additionally considered. However, the associations with functional capacity and disease activity were slightly stronger.

Table 4. Adjusted hazard ratios for developing heart failure de novo*
 Adjusted for age, sex, heart failure, CVD, body mass index, and chronic lung diseaseMultivariate analysis final model
Adjusted HR95% CIPAdjusted HR95% CIP
  • *

    HR = hazard ratio; 95% CI = 95% confidence interval (see Table 1 for other definitions).

  • By Wald's test.

  • Referent was patients without cardiovascular disease (CVD) at study entry.

Characteristics at study entry      
 Age (per 10 years)1.861.14–3.030.0131.741.05–2.890.032
 Male sex1.730.74––5.950.040
 Patients with any other CVD3.061.16–8.050.0243.001.15–7.810.025
 Body mass index (per 5 units)1.501.04–2.170.0321.461.01–2.110.044
 Chronic lung disease1.550.53–4.580.43   
 Disease duration (per 5 years)1.180.99–1.390.058   
 Rheumatoid factor positive1.830.55–6.140.33   
 Rheumatoid nodules1.390.59–3.270.46   
 Erosive disease1.980.59–6.710.27   
 DAS28 (per 1 unit)1.531.07–2.190.019   
 FFbH (per 10-unit improvement)0.650.53–0.80<0.00010.700.57–0.860.0008
 Propensity score (per 0.1 unit)1.561.15–2.120.0045   
Characteristics at followup      
 DAS28 (per 1 unit)1.701.25–2.300.00071.471.07–2.020.019
  15–<30 mg/liter vs. <15 mg/liter0.970.31–2.980.70   
  ≥30 mg/liter vs. 15 mg/liter2.710.80–9.090.11   
  5–9 mg/liter vs. <5 mg/day1.140.48–2.730.77   
  ≥10 mg/day vs. <5 mg/day1.570.46–5.350.48   
 COX-2 inhibitors2.771.14–6.700.024   
 Anti-TNF vs. conventional DMARDs2.190.90–5.330.0831.660.67–4.100.28

When the influences of age, sex, prevalent cardiovascular disease, body mass index, functional capacity, and disease activity at followup were held constant, the contribution (HR) of anti-TNF agents to heart failure risk was estimated to be 1.66 (95% CI 0.67–4.10) (Table 4). This possible increase in heart failure risk did not achieve statistical significance. Of note, the increase was found only if the DAS28 at followup and the time-varying character of the treatment and the inflammation were taken into account, as required for an effective estimate. Using the same simplified multivariate regression model as shown in Table 3, an HR of 0.70 (95% CI 0.27–1.84, P = 0.47) was calculated for the combined positive and negative effects of anti-TNF agents. In the final multivariate model, no increased risk was found in association with glucocorticoid treatment (P > 0.86). A possible influence remained for COX-2 inhibitors. After adjustment for the parameters shown in Table 4, the HR for COX-2 inhibitors was 2.19 (95% CI 0.90–5.36, P = 0.08).

Risk of worsening of the severity of prevalent heart failure.

In 5 of the 12 patients with prevalent heart failure in whom this disease worsened (41.7%), the mean prednisolone dosage per patient from all followup visits with available data exceeded 10 mg/day. This was significantly higher than the percentage of patients whose mean prednisolone dosage exceeded 10 mg/day among the 86 patients with prevalent heart failure that did not worsen (16.3%). Treatment with high-dose glucocorticoids was also found to be a predictor of worsening of heart failure by multivariate Cox proportional hazards regression analysis. Only male sex also remained significant in this analysis. No increase in the hazard risk was found for treatment with TNFα inhibitors (Table 5).

Table 5. Adjusted hazard ratios for worsening in the 98 patients with prevalent heart failure*
 Adjusted HR95% CIP
  • *

    HR = hazard ratio; 95% CI = 95% confidence interval (see Table 1 for other definitions).

  • By Wald's test.

Characteristics at study entry   
 Age (per 10 years)1.520.56–4.140.42
Characteristics at followup   
 Glucocorticoids, ≥10 mg/day  vs.<10 mg/day3.621.06–12.380.041
 Anti-TNF vs. conventional  DMARDs1.180.30–4.730.81


As expected, significantly higher rates of heart failure were observed in patients in this study who had cardiovascular disorders. This result and the finding of increased heart failure risk with increasing age, male sex, or obesity are consistent with results reported by others (30–32). Furthermore, elevated CRP levels and elevated plasma TNFα levels are known to contribute to disease progression in RA and heart failure. In prospective cohort studies of subjects from the general population, CRP level was found to be a strong and independent predictor of cardiovascular disease and congestive heart failure (4, 33–36). Interestingly, we found that the DAS28 as a composite measure of RA activity was a more stringent predictor of heart failure in RA, whereas the CRP level alone was less specific and its hazard ratio did not reach statistical significance. Furthermore, the association between the DAS28 score and heart failure risk was higher for the DAS28 score before the onset of heart failure than for the DAS28 score before the start of treatment. Similarly, the propensity score, as a composite index of the severity of RA, did not reach statistical significance after adjustment for DAS28 at followup.

The results of multivariate Cox regression analysis were supported by the findings of a small nested case–control study. Selecting controls who were similar to the incident heart failure cases with respect to non–RA-specific risk factors, we found that these controls had clinically relevant and statistically significantly lower DAS28 scores at followup compared with the heart failure cases. In addition to the degree of activity of the rheumatic disease, low functional capacity (FFbH) was found to be a significant predictor of heart failure.

Our data therefore suggest that controlling the inflammatory activity of RA not only leads to a better outcome of the rheumatic disorder, but also contributes to a reduction of cardiovascular risk. This is consistent with previous findings (8, 37–40). TNFα inhibitor or DMARD treatment that effectively reduces inflammatory activity thus reduces cardiovascular risk as well (41). We therefore included the DAS28 at followup in our final multivariate model, to estimate whether there was any further risk caused by treatment with anti-TNF agents, glucocorticoids, or COX-2 inhibitors. Additionally taking into account the time-varying character of the treatment and the inflammation, it was not possible to rule out the possibility of a residual heart failure risk caused by TNFα inhibition. Although 4,248 patients were included in this analysis, the number of heart failure cases was not high enough to detect or to rule out an increase in the HR of, e.g., 1.5 with sufficient power of 80%. To do so, the sample size or followup time would have had to be ∼10 times greater (42).

Nevertheless we also investigated an overall impact of anti-TNF treatment, as a combined result of a likely preventive effect on heart failure risk if the anti-TNF treatment was effective, and a possible harmful effect of the drugs themselves. Our data suggest that the preventive effect is probably larger than the harmful effect.

Our findings require further validation, especially with respect to worsening of heart failure but also with respect to development of heart failure de novo. We were not able to apply a standardized procedure for diagnosing heart failure, but had to rely on diagnoses made by hospital-based and private practice internists. Nevertheless, our findings are consistent with those of 3 large randomized controlled trials in patients with congestive heart failure. Treatment with etanercept or infliximab (5 mg/kg body weight) did not result in higher mortality or hospitalization rates than occurred with placebo treatment (17, 18, 43). Only the 10 mg/kg infliximab treatment was associated with higher rates (18). However, infliximab is not prescribed at this high dosage for our RA patients in Germany (44).

Glucocorticoids contribute to increased blood pressure, insulin resistance, visceral adiposity, accelerated atherosclerosis, and endothelial dysfunction (8, 45, 46). These effects are harmful to the cardiovascular system. Glucocorticoids are therefore considered risk factors for heart failure. However, other effects of these agents, such as suppression of inflammation or cellular proliferation, may be beneficial (8, 45). Distinguishing such positive and negative effects of glucocorticoids was beyond the scope of the present study, but our data suggest that their contribution to heart failure risk will be overestimated if one does not control for the inflammatory activity of the underlying disease.

Randomized trials have shown increased cardiovascular risks associated with selective COX-2 inhibitors compared with placebo treatment (47–49). These risks differ by drug and dosage (50), but include risk of heart failure (48, 51, 52). Although we were not able to investigate this in detail, we included prescription of COX-2 inhibitors in our models to adjust for possible confounding. We found some further, but weak, evidence of cardiovascular risk associated with these agents.

In summary, patients with severe rheumatoid arthritis, especially those with highly active disease, are at increased risk of developing heart failure. Treatment with COX-2 inhibitors or glucocorticoids may further contribute to this risk. If treatment with TNFα inhibitors is effective in reducing the inflammatory activity of the rheumatic disorder, it is more likely to be beneficial than harmful with regard to the risk of heart failure. Screening for cardiac risk factors and effective treatment of both the rheumatic disorder and the cardiac disease are essential.


Dr. Listing 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 design. Listing, Zink.

Acquisition of data. Kekow, Schneider, Kapelle, Wassenberg.

Analysis and interpretation of data. Listing, Strangfeld, Schneider, Zink.

Manuscript preparation. Listing, Strangfeld, Kekow, Schneider, Kapelle, Wassenberg, Zink.

Statistical analysis. Listing, Strangfeld.

Coding and classifying of serious adverse events. Strangfeld.


Essex, Wyeth, Amgen, and Abbott received the manuscript 30 days prior to submission but did not have any influence on the design and conduct of the study, data collection, development of the analysis plan, or preparation and conduct of the analysis. The interpretation of the data and the drafting, critical revision, and approval of the final manuscript were done independently from the sponsors.


The authors would like to thank the following rheumatologists (all in Germany) who each enrolled at least 25 patients in the study: Ulrich von Hinüber, MD (Hildesheim); Winfried Demary, MD (Hildesheim); Andreas Krause, MD (Immanuel Hospital, Berlin); Maria Stoyanova-Scholz, MD (Wedau Kliniken, Duisburg); Karin Babinsky, MD (Halle); Thilo Klopsch, MD (Neubrandenburg); Rainer Dockhorn, MD (Weener); Constanze Richter, MD (Stuttgart); Gerd-Rüdiger Burmester, MD (Charité University School of Medicine, Berlin); Karin Rockwitz, MD (Goslar); Arnold Bussmann, MD (Geilenkirchen); Hans Peter Tony, MD (Medizinische Poliklinik der Universität Würzburg, Würzburg); Katja Richter, MD (Universitätsklinikum Carl Gustav Carus, Dresden); Brigitte Krummel-Lorenz, MD (Frankfurt/Main); Anett Grässler, MD (Pirna); Elke Wilden, MD (Cologne); Michael Hammer, MD (St. Josef-Stift, Sendenhorst); Edmund Edelmann, MD (Bad Aibling); Christina Eisterhues, MD (Braunschweig); Wolfgang Ochs, MD (Bayreuth); Thomas Karger, MD (Eduardus-Krankenhaus, Cologne-Deutz); Michael Bäuerle, MD (University of Erlangen, Erlangen); Herbert Kellner, MD (Munich); Silke Zinke, MD (Berlin); Angela Gause, MD (Elmshorn); Lothar Meier, MD (Hofheim); Karl Alliger, MD (Zwiesel); Martin Bohl-Bühler, MD (Brandenburg); Carsten Stille, MD (Hannover); Susanna Späthling-Mestekemper, MD (Munich); Thomas Dexel, MD (Munich); Peter Herzer, MD (Munich); Harald Tremel, MD (Hamburg); Stefan Schewe, MD (Medizinische Poliklinik der Ludwig-Maximilians-Universität, Munich); Helmut Sörensen, MD (Krankenhaus Waldfriede, Berlin); Florian Schuch, MD (Erlangen); Klaus Krüger, MD (Munich); Andreas Teipel, MD (Leverkusen); Kirsten Karberg, MD (Berlin); Gisela Maerker-Alzer, MD (Holzweiler); Dorothea Pick, MD (Holzweiler); Volker Petersen, MD (Hamburg); Kerstin Weiss, MD (Lichtenstein); Werner Liman, MD (Ev. Krankenhaus, Hagen-Haspe); Kurt Gräfenstein, MD (Johanniter-Krankenhaus im Fläming, Treuenbrietzen); Jochen Walter, MD (Rendsburg); Werner A. Biewer, MD (Saarbrücken); Roland Haux, MD (Berlin); Wolfgang Gross, MD (Lübeck); Michael Zänker, MD (Evangelisches Freikirchliches Krankenhaus, Eberswalde); Gerhard Fliedner, MD (Osnabrück); Thomas Grebe, MD (Ev. Krankenhaus, Kredenbach); Karin Leumann, MD (Riesa); Jörg-Andres Rump, MD (Freiburg); Joachim Gutfleisch, MD (Biberbach); Michael Schwarz-Eywill, MD (Evangelisches Krankenhaus, Oldenburg); Kathrin Fischer, MD (Greifswald); Monika Antons, MD (Cologne).

The authors are also grateful to Franka Hierse for statistical support and to Ulrike Kamenz and Christina Bungartz for their careful monitoring of the study.