Background While heart failure (HF) is associated with elevations in tumor necrosis factor (TNF)α, several trials of TNF antagonists showed no benefit and possibly worsening of disease in those with known severe HF. We studied the risk of new or recurrent HF among a group of patients receiving these agents to treat rheumatoid arthritis (RA).
Methods We used data from four different US healthcare programmes. Subjects with RA receiving methotrexate were eligible to enter the study cohort if they added or switched to a TNF antagonist or another non-biological disease modifying antirheumatic drug (nbDMARD). These groups were compared in Cox regression models stratified by propensity score decile and adjusted for oral glucocorticoid dosage, prior HF hospitalisations, and the use of loop diuretics.
Results We compared 8656 new users of a nbDMARD with 11 587 new users of a TNF antagonist with similar baseline covariates. The HR for the TNF antagonists compared with nbDMARD was 0.85 (95% CI 0.63 to 1.14). The HR was also not elevated in subjects with a history of HF. But, it was elevated prior to 2002 (HR 2.17, 95% CI 0.45 to 10.50, test for interaction p=0.036). Oral glucocorticoids were associated with a dose-related gradient of HF risk: compared with no use, 1≤5 mg HR 1.30 (95% CI 0.91 to 1.85), ≥5 mg HR 1.54 (95% CI 1.09 to 2.19).
Conclusions TNF antagonists were not associated with a risk of HF hospital admissions compared with nbDMARDs in this RA population.
- Rheumatoid Arthritis
- Cardiovascular Disease
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Studies in animals and humans demonstrate elevations in tumor necrosis factor (TNF)α associated with heart failure (HF) and left ventricular dysfunction.1–3 TNFα is released by myocytes under conditions of stress4 and plays a role in obstructive coronary artery disease.5 Elevations in TNFα in the setting of HF suggested a therapeutic role for TNF blockade. While TNF antagonists were approved for their use in rheumatoid arthritis (RA), several large scale trials tested whether they might improve HF symptoms among patients without inflammatory or autoimmune diseases.6 ,7 Neither of the trials—one with etanercept and another with infliximab—found a beneficial effect of TNF antagonists; in fact, there was evidence that subjects with severe HF were harmed by these agents in one such study.6
These results and a case series from the Food and Drug Administration Medwatch program8 raised substantial concern about the use of TNF antagonists among patients with RA. No clear signal for HF risk was observed in the initial TNF antagonist trials for RA; however these trials were relatively short and inadequately powered to observe excess HF risk. Observational studies have produced inconsistent results, some suggesting a reduced risk of HF, others an increased risk and others no difference in risk.9–12 These studies used various methods, all with well-described limitations: some did not focus on new users of TNF antagonists, some compared TNF antagonist users with non-users, others did not adjust for cardiovascular risk factors, and none examined many of the subgroups of interest, such as oral glucocorticoid and non-steroidal anti-inflammatory drug (NSAID) users.
We attempted to overcome some of the limitations of prior studies by studying the risk of HF among a large group of patients with RA starting a TNF antagonist or a non-biological disease modifying antirheumatic drug (nbDMARD).
Design and Study Cohort: These analyses were part of a larger study collaborative (The Safety Assessment of Biologic thERapy) for which many of the methods have previously been described in detail.13 Briefly, limited de-identified datasets obtained by the collaborators were shared across institutions to facilitate large-scale studies of comparative safety and effectiveness. The study combined information from the US Medicaid Analytic Extract linked to national US Medicare data for people insured by (so-called ‘dual eligibles’), the Tennessee Medicaid file (TennCare), two US states’ Medicare population databases, and Kaiser-Permanente of northern California's electronic health record.14 Information contained in these databases includes limited sociodemographic data (age, gender, race), enrolment dates, inpatient and outpatient healthcare encounter insurance claims with diagnoses (ICD-9-CM) and procedure codes (Current Procedural Terminology) and all pharmacy claims. The datasets included information from 1998–2007.
From the potentially eligible study population, we selected persons with at least one encounter associated with a diagnosis of RA (ICD 714, excluding 714.3) who were over 16 years of age at the RA diagnosis date. We excluded patients with a diagnosis of ankylosing spondylitis or psoriatic arthritis. To improve the consistency of baseline disease severity across the study population, we required the cohort to have been users of methotrexate (oral or injectable) immediately prior to the start of follow-up. Requiring a diagnosis and a prescription for a disease-modifying antirheumatic drug has high positive predictive value for RA.15 Follow-up began if and when the subject added or switched to an available TNFα blocking agent (adalimumab, etanercept or infliximab; certolizumab pegol and golimumab were not yet available) or a non-methotrexate nbDMARD (hydroxychloroquine, leflunomide or sulfasalazine).
The study protocol was reviewed and approved by the responsible Institutional Review Boards.
TNFα Antagonists and Non-biological DMARD Exposures: Subjects were followed from the date of switching from methotrexate to a TNFα antagonist or an nbDMARD, or adding one of those drugs to their methotrexate regimen; these dates were termed the index dates. We prespecified two exposure definitions: (1) the primary exposure definition considered subjects in their first treatment group as long as prescriptions continued for that regimen (‘as treated’), and (2) the secondary exposure definition considered subjects in the first treatment group throughout only the first 6 months of follow-up regardless of any changes in treatment (‘first exposure carried forward’). In the as-treated analysis, follow-up ended at the first of any of the following occurrences: on the date the specified treatment was stopped plus 30 days, on the date when a TNFα antagonist was added to or substituted for the non-methotrexte nbDMARD, on the date of a HF outcome or death. Subjects could switch agents within a class of drugs (ie, between nbDMARDs or TNFα antagonists). We allowed subjects in the TNFα blocking agent group to concurrently use nbDMARDs but not the reverse.
HF Outcomes: While potentially related, these analyses focused on HF hospitalisations and not ischaemic heart disease. Acute care hospitalisations with the diagnosis of HF (ICD 428.x) in the primary or principal position have been found to accurately identify patients with HF admissions; several prior studies demonstrate a positive predictive value >80%.16 ,17 We considered the composite of new or recurrent HF as the primary endpoint and each component as exploratory secondary outcomes. New and recurrent HF were based on the absence or presence of HF diagnosis codes from inpatient or outpatient visits during the 365 day baseline period prior to the start of follow-up. An additional secondary outcome explored was the presence of HF in any position (not just the primary position) on the acute care discharge claim.
Potential Confounders: We used propensity score methods to reduce confounding. Generally, a propensity score estimates the probability of receiving one treatment versus another and is a summary score that can adjust for confounding in place of individual covariates.18 We used multivariate logistic regression models to estimate the propensity score, which we defined as the predicted probability of a patient receiving a TNFα antagonist versus a nbDMARD. The propensity score model included confounders that can be measured using administrative data (demographics, diagnoses, surgical procedures and pharmacy dispensings; see online supplementary table for a listing of variables). These included prior cardiovascular diagnoses, cardiovascular risk factors and cardiovascular medication use. These variables were determined over the 365 days before the start of follow-up and were derived for the main study population; this same propensity score was applied to each of the subgroups of interest (eg, known HF patients). For double robustness, several variables of special interest were included in the propensity score and as individual covariates in the outcome (HF) model, such as number of HF hospitalisations and the use of loop diuretics in the 365 days prior to the start of follow-up. The use of oral glucocorticoids was also included as an individual covariate; we calculated the cumulative prescribed dosage in prednisone equivalents during the 180 days before the start of follow-up.
Statistical Analyses: In the primary analyses, subjects were split into deciles of propensity scores. Subjects within a decile should have approximately equal distributions of the component covariates of the propensity score; to maximise comparability of patients within each decile, we excluded (‘trimmed’) subjects with the top and bottom 5% of propensity scores from the combined cohort.19 We examined the distribution of baseline characteristics in both exposure groups.
Incidence rates of the HF outcomes with 95% CI were calculated for each exposure group separately. We constructed Cox regression models comparing the risk of the composite HF outcome among those exposed to a TNFα antagonist or nbDMARD, with the censoring described above. The models were stratified by decile of propensity score, and adjusted for 180-day prior cumulative oral glucocorticoid dosage, the number of HF hospitalisations and the use of a loop diuretic. In our secondary analyses, we used the same outcome models to examine the risk of new and recurrent HF.
Sensitivity analyses tested whether the results depended on the exposure definition and whether results were robust across subgroups. We applied the secondary exposure definition, first exposure carried forward, in Cox regression models similar to the primary analysis. We also extended the primary exposure definition, as treated, to 12 months of follow-up.
Then, we examined relevant subgroups, using the primary exposure definition. Subgroups based on data from the 1 year prior to the start of follow-up included: age ≥65 or <65 years, gender, prior cardiovascular disease or diabetes, prior statin use, prior non-selective or selective NSAID use, prior acute care hospitalisations for HF and the use of loop diuretics, and start of follow-up prior to or after 2002. This last subgroup analysis was pursued because a warning was inserted into the labelling of TNF antagonists regarding HF in late 2001.
The proportional hazards assumption was tested using the Kolmogorov supremum test of Lin, Wei and Ying and was not violated in the primary analyses (p=0.10).20
After combining datasets, we found 139 611 potentially eligible patients with a diagnosis of RA. Requiring use of methotrexate and addition or switch to a TNF antagonist or another nbDMARD excluded106 704. From the remaining 22 907 eligible subjects, 2664 were excluded by trimming the top and bottom 5% based on propensity scores, leaving 20 243 subjects for these analyses.
The TNF antagonist group and nbDMARD group were similar after trimming (see table 1). The trimmed groups had a mean age of 56 years, 86% women and were mostly Caucasian. Similar percentages of subjects in both treatment groups had experienced a prior myocardial infarction (MI) (1.8–1.9%), stroke (1.7%) or coronary revascularisation (0.7%). As well, cardiovascular risk factors were similar: diabetes (22.9–23.5%), hypertension (41.6–42.2%) and hyperlipidaemia (52.3–53.5%). Prior HF was slightly more common in subjects who used an nbDMARD (4.2%) as compared with a TNF antagonist (3.7%). Loop diuretic use (14.8–15.0%) and ACE inhibitor use (23.2–23.3%) was very similar across groups. The group with no known HF was largely similar to the total cohort, and the prior HF group differed from the no known HF group in expected ways, but was similar across treatment arms (see table 1).
The incidence rates (per 100 person-years) for HF are shown in table 2. The group receiving TNF antagonists appeared to have a slightly lower incidence rate of HF in the total population (2.49 per 100 person-years) than that receiving nbDMARDs (3.15), but the CIs are widely overlapping. The rates of HF among subjects with a history of HF (26.7–32.2) were 10-fold to 15-fold greater than the group without a known history of HF (1.65–1.95).
We calculated HR in Cox regression models to compare the risk of HF in the TNF antagonists to nbDMARD users (see table 3). There was no significant increase in HF risk associated with TNF antagonist use in either the primary analysis (HR 0.85, 95% CI 0.63 to 1.14) or secondary (HR 0.90, 95% CI 0.69 to 1.16) analyses. These results were robust across the two secondary outcomes, new HF and recurrent HF. In the primary analysis, oral glucocorticoid use was associated with a dose-dependent risk of HF: compared with no use, 1–4.9 mg HR 1.30 (95% CI 0.91 to 1.85), 5+ mg HR 1.54 (95% CI 1.09 to 2.19). However, use of oral glucocorticoids did not modify the effect of TNF antagonists on HF risk (see figure 1).
Sensitivity analyses examined the HRs for HF across relevant subgroups (see figure 1). No subgroup analyses showed results significantly different than the main analysis. However, men using TNF antagonists (HR 2.23, 95% CI 0.78 to 6.35) and the group initiating TNF antagonists before 2002 (HR 2.17, 95% CI 0.45 to 10.50) had numerically elevated risks of HF. The gender by treatment interaction analysis was not significant (p value=0.40), but the interaction term for time period (pre-2002 or later) by treatment was statistically significant (p value=0.036).
We examined the strongest HF risk factors in the TNF antagonist versus nbDMARD groups before 2002 versus 2002 or later. Differences in history of HF, MI and use of oral glucocorticoids were modest, but show a higher prevalence of HF and MI and higher use of glucocorticoids in the TNF antagonist group pre-2002. In the later time period, the prevalence of HF was modestly higher in the nbDMARD group and MI and glucocorticoid differences were less marked (see table 4).
An additional secondary analysis allowed the HF diagnosis in any position in the acute care hospital discharge claim. The results were very similar to the primary analysis (HR 0.84, 95% CI 0.70 to 1.02).
In randomised controlled trials, the use of TNF antagonists for the treatment of HF was not successful; these trials suggested possible harmful effects for persons with severe HF.6 ,7 Prior studies confused providers and RA patients regarding the safety of TNF antagonists. Past observational studies in RA patients demonstrated mixed results—some suggested TNF antagonists may be associated with increased HF risk10 and others found no increased risk,12 or a reduced risk.9 ,11 In this large study of subjects with RA enrolled in various healthcare benefits programmes, we found little evidence of HF risk associated with TNF antagonists. The absence of risk was observed using various exposure definitions, different outcomes (new and recurrent HF), and among several relevant subgroups. We found dose-related risk of HF among users of oral glucocorticoids, agreeing with prior literature.21 ,22
In October 2001, the Food and Drug Administration issued a black box warning regarding HF and TNF antagonists. It is possible that this warning and/or other information about potential risks of TNF antagonists caused a change in prescribing practices for TNF antagonists, such that patients at highest risk of HF and/or with more severe HF were less likely to be prescribed these medications. Because of this concern, we performed an analysis stratifying by time period (pre-2002 and later). Although TNF antagonists were not associated with HF in either time period, the HR for this association were in the opposite direction, suggesting that compared with nbDMARDs, TNF antagonists were associated with a higher risk in the earlier time period and a lower risk in the later time period. We thus compared the strongest risk factors for HF in the two comparison groups after stratifying by time period. Compared with nbDMARD users, TNF antagonist users had a higher prevalence of HF, myocardial infarction, and glucocorticoid use prior to 2002. In the later time period, TNF antagonist users had a lower prevalence of HF and other risk factors were more similar between the two groups (see table 4). Although our analysis controlled for known risk factors for HF including those in table 4, it is possible that other HF risk factors were not measured and not accounted for in our analyses. Such residual imbalances, if not controlled for, could result in the modest differences in HR we observed when stratifying by time period.
Strengths of the current study are worth noting. It was relatively large in size from a nationwide sample, allowing for relevant subgroup analyses. We used a new user analysis and tested several definitions of exposure.23 Drawing subjects from a source population of methotrexate users and comparing TNF antagonists to an active comparator group of new nbDMARD users helps to reduce unmeasured confounding.24
Limitations of our methods require discussion. The study dataset did not contain any validated markers of RA disease severity. The cohorts appeared well balanced with respect to most measured covariates, however it is possible that one of the two groups had more severe RA that might have predisposed toward more HF. While disease severity might be higher in patients selected for a TNF antagonist, concerns regarding a possible association between these agents and HF might have limited their use. Furthermore, the groups may not have been balanced with respect to HF risk factors or left ventricular function, as discussed above. While table 1 suggests overall balance across many factors (including prior HF, loop diuretic use and ACE inhibitor use), an insurance claims database has limited information about HF risk factors and contains no laboratory results. It is possible that some information goes unrecorded in an insurance claims database (‘missing data’), but it is impossible to determine the extent of unrecorded information. Patients are not considered lost to follow-up in an insurance claims database because medical and pharmacy bills are submitted if services are rendered. There is likely some misclassification of RA and the HF endpoint; however, it is not likely to be substantial nor differential based on prior work validating these algorithms.15–17 It is possible that a longer ascertainment period for covariates would have yielded increased rates of certain variables, however this would have come at the expense of a shortened follow-up period and a smaller cohort. We found relatively high rates of oral glucocorticoids during the baseline period (75–80%), but this did not differ by exposure. In addition, there is likely misclassification of the known prior HF and no known prior HF groups, since our data do not include a subject's full history. However, it is interesting to note that the rates of HF in subjects without known HF in our study were almost identical to rates observed in a population-based cohort of RA from Olmstead County (1.65–1.95 per 100 person-years in our study vs 1.99 per 100 in Olmstead County).25 The external validity of our study needs further examination because many patients were in Medicaid and/or Medicare. Finally, several potentially important variables were unmeasured, including over-the-counter NSAIDs, smoking, blood pressure, body mass index, physical activity and aspirin use. It is possible that these unmeasured variables are differentially distributed across the exposure groups.
The findings of this study have several clinically relevant implications. First, the results suggest that in this cohort of RA, TNF antagonists were not associated with an increased risk of HF compared with nbDMARDs. Our study does not rule out the possibility of an increased risk of HF among certain subgroups, such as those with structural heart disease. However, no suggestion of increased risk was observed in the high risk group of patients with a history of HF. Second, whenever possible, the use of oral glucocorticoids should be limited; our data suggests that small reductions in dosage may ameliorate the risk of HF. Third, our findings regarding oral glucocorticoids suggest that studies of HF in RA, including clinical trials, need to carefully control for this important comedication. There is an increasing literature regarding an elevated HF risk among patients with RA, suggesting worsened left ventricular dysfunction and a higher risk of diastolic dysfunction.26 ,27 Such studies must adequately control for medication effects. Several prior observational studies in the general population suggested that TNF antagonists improved left ventricular ejection fraction.28 ,29 However, the subsequent negative trial results point out the importance of testing such strategies in randomised placebo controlled trials.
In conclusion, we studied the risk of HF among patients with RA starting a TNF antagonist or a nbDMARD after use of methotrexate. We found no elevation in HF risk among TNF antagonist users. This was the same for those with and without known prior HF. We did find a significant elevation in HF risk with higher dosages of oral glucocorticoids, consistent with prior literature. Since the literature suggests that RA is associated with HF, presumably through the effects of inflammation on cardiac contractility, it is critical to better understand the effects of potent immunosuppressives on HF risk.
This work was part of a larger collaborative, the Safety Assessment of Biologic Therapy (SABER). This collaboration includes: Agency for Healthcare Research and Quality, Parivash Nourjah; Brigham and Women's Hospital, Robert Glynn, Mary Kowal, JL, JAR, Sebastian Schneeweiss, DHS; Meyers Primary Care Institute and University of Massachusetts, LRH; Food and Drug Administration, Jane Gilbert, DJG, Carolyn McCloskey, Rita Ouellet-Hellstrom, Kristin Phucas, James Williams; Kaiser Permanente Northern California, Lisa Herrinton, LL, Marcia Raebel; Kaiser Permanente Colorado, Marcia Raebel; University of Alabama at Birmingham, LC, JRC, Elizabeth Delzell, Nivedita Patkar, Kenneth Saag, Fenglong Xie; University of Pennsylvania, Kevin Haynes, JDL, Vanderbilt University, MRG, Carlos Grijalva, Ed Mitchel. We are indebted to the Tennessee Bureau of TennCare of the Department of Finance and Administration, which provided the TennCare data.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online table
Handling editor Tore K Kvien
Contributors Study Design: DHS, JAR, DJG, JDL, MRG, JRC. Data Acquisition: DHS, MRG, LL, JRC. Analysis: DHS, JAR, JL, LC, JDL, JRC. Drafting manuscript: DHS, JAR, BK, JDL, MRG, JRC. Final approval of manuscript: DHS, JAR, BK, LC, LRH, DJG, JDL, JL, LL, MRG, JRC.
Funding This work was supported by the Agency for Healthcare Research and Quality (AHRQ) and the Food and Drug Administration (FDA) US Department of Health and Human Services (DHHS) as part of a grant (No. 1U18 HSO17919-0) administered through the AHRQ CERTs Program. The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by AHRQ, FDA or DHHS.
Competing interests DHS has received research grants from Abbott, Amgen and Lilly, has served in unpaid roles on two Pfizer trials not related to rheumatoid arthritis, has directed an educational course supported by Bristol Myers Squibb, and serves as a consultant to CORRONA. JRC has received research grants and/or consulting with Amgen, Abbott, BMS, Genentech/Roche, Janssen, UCB, and CORRONA. LRH serves as a consultant to CORRONA. JDL has received research grant from Centocor and served as a paid consultant to Amgen and Pfizer.
Ethics approval Partners Healthcare Human Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.