|Home | About | Journals | Submit | Contact Us | Français|
New onset heart failure (HF) has been associated with the use of TNF-α antagonists etanercept and infliximab based upon spontaneous adverse event reports. HF clinical trials of these agents were stopped early due to futility or worsening of existing HF. A potential association between etanercept and infliximab and new onset HF has been studied minimally at a population level.
Using administrative claims from a large U.S. health care organization, we identified rheumatoid arthritis (RA) and Crohn’s disease (CD) patients receiving infliximab or etanercept (exposed), and comparator cohorts of RA and CD patients receiving non-biologic immunosuppressives (unexposed). We studied adults <50 years to reduce potential confounding related to common age-related comorbidities. Based on abstracted medical records of suspected HF cases, a physician panel adjudicated cases as definite, possible or no HF.
Among 4018 RA and CD patients with mean duration follow-up of 18 months, 9 of 33 suspected HF cases (identified using claims data) were adjudicated as definite (n = 5) or possible (n = 4) HF. The relative risk of HF among TNF-α antagonist-treated RA and CD patients was 4.3 and 1.2, respectively (P = NS for both). The absolute difference in cumulative incidence of HF among infliximab or etanercept-exposed compared to unexposed patients was 3.4 and 0.3 cases per 1000 persons for RA and CD (P = NS), respectively, yielding a number needed to harm of 294 for RA and 3333 for CD.
We found only a small number of presumed HF cases (n = 9, or 0.2%) in a large population of relatively young RA and CD patients. Although there was an increased relative risk of incident, HF that was not statistically significant among those exposed to TNF-α antagonists compared to those unexposed, larger cohorts are needed to provide more precise risk estimates and permit adjustment for potential confounding.
Tumour necrosis factor alpha (TNF-α) antagonists have emerged as an effective therapy for patients with rheumatoid arthritis (RA), Crohn’s disease (CD) and other conditions. Early data suggested they might be a promising treatment for heart failure (HF). However, randomized controlled trials of TNF-α antagonists in HF patients were halted due to lack of benefit or trends towards worsened HF in patients receiving higher TNF-α antagonist doses . Based on reports submitted to the FDA Adverse Event Reporting System, a series was subsequently published describing 47 cases of new or worsening HF in patients who had received TNF-α antagonists . Nine of the patients in this case series had an exacerbation of pre-existing HF, and of the 38 patients with new onset HF, half had no cardiovascular risk factors for HF; 10 patients in the case series were under age 50 yrs. However, data from spontaneous case reports can provide neither incidence rates nor estimates of risk compared to an unexposed but comparable population. Thus, the data in this case series were insufficient to allow inferences with regard to causal associations. Subsequent studies quantifying the risk of HF in TNF-α antagonist-treated patients have not focused on this association in younger individuals who are generally at low risk for HF based on age .
To investigate a possible association between TNF-α antagonist use and incident HF, we examined a cohort of TNF-α antagonist-exposed RA and CD patients, and determined the incidence and relative risk of new onset HF compared to a population of RA and CD patients not receiving these agents. We specifically focused our attention on patients younger than 50 years of age given their expected low prevalence of age-related cardiovascular comorbidities and risk factors.
After institutional review board approval, we utilized the medical and pharmacy administrative claims from a large geographically diverse U.S. health care organization with membership in more than 15 states from January 1998 to December 2002 to identify RA and CD patients younger than age 50 years . For each patient, we required at least two ICD9-CM diagnosis codes for RA (714.X) or CD (555.X) during the study period and also required that each individual had received an infusion or filled a prescription for a TNF-α antagonist (i.e. etanercept or infliximab) or filled at least three prescriptions for one of several selected immunosuppressive drugs. TNF-α antagonist users were considered the exposed cohort. Comparator (TNF-α unexposed) RA patients filled ≥3 prescriptions for methotrexate (MTX), and comparator (TNF-α unexposed) CD patients filled ≥3 prescriptions for MTX, 6-mercaptopurine, azathioprine or prednisone >10 mg/day. We required that the unexposed cohort fill prescriptions for the medications indicated to select comparator patients receiving non-biologic therapies commonly used to treat RA and CD. Although these diagnosis and medication requirements may have excluded patients with early or mild disease, we required these criteria in order to study relatively homogeneous cohorts that had a pattern of claims data suggestive of greater certainty in disease diagnosis. For all glucocorticoids other than prednisone, we converted dosages to prednisone-equivalent dosages.
Date of first exposure to the TNF-α antagonist or the third dispensing of the comparator drug defined each person’s ‘index date’. Potential confounders and covariates of interest were examined in the administrative data in the 6 months prior to each member’s index date. Individuals with a diagnosis of HIV disease, organ transplantation or malignancy in the 6 months prior to the index date were excluded from the study. Persons with a claims-identified diagnosis of HF prior to their index date also were excluded from the study, as we wished to examine only new onset HF.
Among the study population, we identified claims with one or more in-patient or out-patient diagnosis codes for heart failure (ICD9-CM 428.xx, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 425.4, 425.5, 425.7, 425.8, 425.9) on any type of medical claim or diagnostic test following the index date. To maintain a biologically plausible link to the medication exposure, we limited our investigation of heart failure claims to those involving services occurring within 9 months of the most recent exposure to a study drug (TNF-α antagonist or comparator).
Using a pilot-tested data collection tool, experienced nurse abstractors reviewed the medical records of patients with suspected HF identified from the administrative claims data. The data abstracted included modified elements of the Framingham criteria for heart failure  and ejection fraction (EF) (modified major criteria: diagnosis of HF recorded by a physician, shortness of breath or dyspnoea on exertion, orthopnoea or paroxysmal nocturnal dyspnoea, EF < 40%, pulmonary oedema, rales or crackles on chest exam; modified minor criteria: peripheral oedema, pleural effusion, cardiomegaly or one of the following medications: angiotensin converting enzyme inhibitors, angiotensin receptor blockers, hydralazine, nitrates, loop diuretic, digoxin, beta-blocker, dobutamine, nesiritide, spironolactone). The medical records of cases that satisfied at least one major or two minor modified Framingham criteria were subsequently reviewed independently by two physicians (including a cardiologist) to determine by clinical judgment whether the case represented ‘definite’, ‘possible’ or ‘no’ heart failure. Modification of the Framingham criteria to include EF < 40% as a screening criterion was adopted in light of the previous case series from the Food and Drug Administration’s Adverse Event Reporting System (AERS) that suggested the pattern of the incident severe heart failure in the younger patients without comorbidities receiving TNF-α antagonists involved systolic dysfunction . The reviewing physicians did not assess the cause of HF or require that the EF be <40% but only determined whether HF was present and of new onset. One reviewer was blinded to study objectives and design. Although the intent was to completely blind both reviewers to TNF-α antagonist exposure status, mention of TNF-α antagonist use was included in some of the medical records that were reviewed. Initially discordant assessments were resolved by discussion and consensus.
We used cumulative incidence ratios to compare the risk of physician-assessed ‘definite’ or ‘possible’ incident heart failure between the TNF-α exposed cohort and the unexposed cohort. We also examined the risk stratified by disease indication (RA or CD) and by the specific TNF-α antagonist used (among the exposed). Exact methods were used to compute confidence intervals for cumulative incidence estimates . Confidence intervals for the cummulative incidence ratio comparing TNF-α antagonist exposed vs unexposed were derived using the normal approximation . In recognition that the normal approximation for confidence intervals is marginally valid when comparing groups with small numbers of cases (e.g. less than five), we did not compute confidence intervals for subgroup analyses. Analyses were performed using SAS 9.1 (SAS Institute, Cary, NC, USA).
A total of 2121 patients with RA and 1897 with Crohn’s disease younger than age 50 years were identified in the study population. Their demographic and health service utilization characteristics during the study period are shown in Table 1. The mean age and gender composition of the TNF-α exposed and unexposed groups were similar. Persons in the TNF-α antagonist exposed group had more physician visits and, in the RA cohort, were more likely to have been hospitalized. RA and CD patients exposed to the TNF-α antagonists had more extra-articular or extra-intestinal involvement and morbidity requiring intervention (e.g. surgery). Pre-existing cardiovascular disease diagnoses were uncommon and their prevalence was similar between exposed and unexposed groups. Less than 10% of each of the cohorts had prior exposure to the various classes of cardiovascular medications. Approximately half of the RA cohort and less than one-fourth of the CD cohort were glucocorticoid users.
Following abstraction of 29 medical records of suspected heart failure cases (0.7% of the entire cohort), 20 (0.5%) met sufficient modified Framingham criteria to warrant physician review of the abstracted medical records (an additional four records were requested but not received). After review of the medical records, a total of nine cases (0.2%) were determined by the physician reviewers to be ‘definite’ or ‘possible’ HF. Using these results, Table 2 shows the cumulative incidence and relative risk of heart failure among TNF-α exposed patients compared to TNF-α unexposed patients stratified by disease indication. Among RA patients, the cumulative incidence of definite or possible HF among those treated with TNF-α antagonists was 4.4 cases per 1000 persons compared to 1.0 cases per 1000 persons in the unexposed cohort. Based on an absolute risk difference of 3.4 cases per 1000 persons, the number needed to harm (NNH) was 294. For Crohn’s patients, the absolute risk difference was even smaller (0.3 cases per 1000 persons) and the NNH larger (3333). There was a statistically non-significant increased crude relative risk of heart failure among all TNF-α exposed RA or CD patients compared to unexposed patients (relative risk 4.3 and 1.2, respectively, P = NS for both). Because of the small number of incident heart failure cases, we were unable to perform multivariable adjustment.
Table 3 shows the clinical characteristics of the definite and possible heart failure cases. Six of the nine cases were in-patients who had received TNF-α inhibitors. Of the three non-exposed cases, two had prior valvular heart disease. The 3rd non-exposed case was a 25-year-old male with documented low EF (35–40%) without valvular abnormality. Of the six cases in TNF-α -exposed patients, one had right-sided HF with associated obesity and sleep-apnoea (case 1, 42-year-old male). One of the ‘definite’ HF cases exposed to TNF-inhibitors had a markedly diminished EF (20–25%) with associated pulmonary oedema, but this patient also had a history of angioplasty and coronary artery bypass graft surgery (case 2, 46-year-old male). Another had pulmonary hypertension presumed related to underlying auto-immune disease (case 7, 34-year-old woman). One ‘possible’ case exposed to TNF-α inhibitors had an EF of 30–35% by transesophageal echo and a recorded diagnosis of viral cardiomyopathy (case 9, 17-year-old female).
We found only a small number (n = 9, or 0.2% of persons) of presumed heart failure cases in a total population of 4018 RA and CD patients. Among these relatively young patients, there was an increase in the relative risk of incident HF that was not statistically significant among patients exposed to TNF-α antagonists compared to patients exposed to other disease modifying agents. Additionally, the clinical details (e.g. interstitial lung disease, valvular abnormalities) of some of the presumed HF cases may have represented an extra-articular feature of the underlying disease process and be unrelated to drug exposure. If patients with these extra-articular disease features were more likely to be treated with TNF-α antagonists, the estimated relative risks would be spuriously increased (confounding by indication). The absolute risk difference of heart failure between the TNF-α antagonist treated RA patients compared to untreated patients was 3.4 cases per 1000 persons, yielding a number needed to harm of 294. Both absolute and relative risk differences for CD patients were negligible.
The risk of heart failure among patients with inflammatory diseases, particularly RA, has been a subject of much discussion in the literature. Recent studies have found that RA patients not exposed to TNF-α antagonists are at increased risk for both incident HF and excess HF-related mortality compared to non-RA patients [7, 8]. Only about half of the excess risk of HF in RA patients can be attributed to traditional cardiovascular risk factors , and both rheumatoid factor positivity and extra-articular RA have been shown to be risk factors for cardiovascular mortality [7, 10]. RA patients may also suffer from disease-related pulmonary and cardiac abnormalities such as interstitial lung disease, secondary pulmonary hypertension and cardiac valvular abnormalities; we observed some of these conditions as described in Table 3. If RA patients at greater risk for heart failure due to their underlying disease are more likely to be prescribed TNF-α antagonists (confounding by indication ), this may account for an observed increased risk of HF in these RA patients. Indeed, studies using only claims data to assess both comorbidities and HF outcomes among RA patients may suffer from a lack of information on important confounders (e.g. body mass index, smoking status, subclinical atherosclerotic vascular disease, valvular disease). In contrast, we are not aware of prior literature suggesting that CD patients are at increased risk for HF (even in the absence of TNF-α antagonism), although unchecked systemic inflammation might play an aetiologic role [10, 12]. Additionally, RA and CD patients treated with TNF-α antagonists could have a higher prevalence of cardiovascular-related comorbidities prior to TNF-α antagonist initiation, although we did not observe this in our relatively young cohort.
Previous studies among TNF-α antagonist-exposed persons with auto-immune diseases found no increase in risk of HF. Wolfe and Michaud  found 42 cases of incident HF in a large cohort of RA patients using HF endpoints defined by a combination of patient self-report via questionnaires, physician contact and medical record review. There was no increased risk of incident HF among TNF-α exposed RA patients compared to those receiving other DMARDs. There were only three cases of incident HF among 2970 patients younger than 50 years (incidence of 1 per 1000), and all were observed in patients receiving only MTX. Concordant with our results, another recent study using administrative claims data showed there was no significant difference in the risk of HF between RA patients exposed to MTX and those exposed to TNF-α antagonists . There was no specific mention of the risk of HF in younger patients. In that study and in contrast to our case validation process, HF endpoints were determined by using an ICD9-CM code for HF without medical record review.
The strengths of our study include a large cohort of RA and CD patients enrolled in a large U.S. health care organization, approximately half of whom were exposed to TNF-α antagonists during a mean observation period of 18 months. Our results address the safety of TNF-α antagonists in this ‘real world’ setting, which is an important adjunct to results from clinical trials that often are limited to highly selected populations. Incident HF outcomes were sought and rigorously assessed using a sensitive, claims-based strategy followed by abstraction of one in-patient or out-patient medical record that was independently reviewed by two physicians. Using this methodology, we found only nine definite or possible cases of incident HF. It is possible that abstraction of additional medical records for each suspected case of HF might have increased the sensitivity of our approach. However, the 33 suspected cases of HF (0.8% of the entire cohort) preliminarily identified using the claims data could be considered a maximal upper bound of the absolute incidence of heart failure in this population; moreover, after medical record review of the 20 cases that met the Framingham screening criteria, it was clear that there was sufficient information to refute the diagnosis of heart failure for some of them. Another notable strength of our work is that we were able to obtain medical records to identify relevant clinical details of the suspected heart failure cases to identify comorbidities (e.g. morbid obesity, longstanding interstitial lung disease with pulmonary hypertension) that are often under-reported in claims data. Other studies that are limited to claims data alone to determine outcomes may suffer from a lack of information about these and other important and often unmeasured factors that may confound exposure–outcome relationships.
Although the overall absolute incidence of heart failure in this population may be reassuring, the increased but non-significant relative risk of heart failure observed among RA patients may be a source of concern. Because of the small number of cases, our study had limited power, and the non-statistically significant increased relative risk we observed might reflect the random play of chance. The small number of HF cases we observed was in part due to our restriction of the study population to persons younger than the age of 50, which was done to focus on a younger population identified in a prior report of HF associated with TNF-α antagonists that included some patients without cardiovascular risk factors . Our study does not address the risk of HF among older persons or those with different patterns of cardiovascular comorbidities. Additionally, the small number of cases precluded adjustment for potentially important confounders of interest. Another limitation of our work was that TNF-α exposure (among treated patients) was mentioned in some of the medical records that were reviewed by physicians that confirmed HF (although these physicians did not have access to the pharmacy claims data).
In conclusion, among younger persons with few cardiovascular comorbidities, we observed only a few presumed cases of incident HF and a relatively low absolute risk, even among those exposed to TNF-α antagonists. Particularly, among RA patients, we did find an increased relative risk associated with TNF-α antagonists that was not statistically significant. Confounding by indication or a chance finding may account, in part or whole, for the increased relative risk estimate we observed. A larger cohort and/or additional follow-up time would provide more precise relative risk estimates and permit adjustment for potential confounding.
We would like to thank Drs Michael Cuffe and Wendy Gattis Stough at Duke University Medical Center for their contributions related to the clinical aspects of heart failure and review of the draft manuscript. We would also like to thank Dr Valerie Ulstad for her work in adjudicating the heart failure endpoints. We thank Aparna Mohan, MD PhD, Nigel Rawson, PhD, for key efforts in the planning and inception of this project, and Kevin Anstrom, PhD, for statistical consultation. Funded by FDA CBER Award #223-02-1420 Task Order #1, the Maryland chapter of the Arthritis Foundation, grant HS10389 from the Agency for Healthcare Research and Quality, K24 AR052361-01 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, and T32 AR47512-03 from the National Institutes of Health. FDA’s Center for Drug Evaluation and Research was not involved in this project.
Disclosure statement: K.G.S has received research grants from Amgen and Centocor. He has served as a consultant with Amgen. J.R.C. has received research grants from Amgen. He has served as a consultant for Roche. All other authors have declared no conflict of interest.
J. R. Curtis, Center for Education and Research on Therapeutics of Musculoskeletal Disorders, The University of Alabama at Birmingham, Birmingham, AL.
J. M. Kramer, Center for Education and Research on Therapeutics of Cardiovascular Diseases, Duke University, Durham, NC.
C. Martin, Center for Health Care Policy and Evaluation, Eden Prairie, MN.
K. G. Saag, Center for Education and Research on Therapeutics of Musculoskeletal Disorders, The University of Alabama at Birmingham, Birmingham, AL.
N. Patkar, Center for Education and Research on Therapeutics of Musculoskeletal Disorders, The University of Alabama at Birmingham, Birmingham, AL.
D. Shatin, Center for Health Care Policy and Evaluation, Eden Prairie, MN.
M. Burgess, Center for Health Care Policy and Evaluation, Eden Prairie, MN.
A. Xie, Center for Education and Research on Therapeutics of Musculoskeletal Disorders, The University of Alabama at Birmingham, Birmingham, AL.
M. M. Braun, Food and Drug Administration, Rockville, MD, USA.