Go AS, Yang J, Gurwitz JH, Hsu J, Lane K, Platt R. Comparative Effectiveness of Different β-Adrenergic Antagonists on Mortality Among Adults With Heart Failure in Clinical Practice. Arch Intern Med. 2008;168(22):2415-2421. doi:10.1001/archinternmed.2008.506
Copyright 2008 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2008
Randomized trials have demonstrated the efficacy of selected β-blockers in systolic heart failure, but the comparative effectiveness of different β-blockers in practice is poorly understood.
We compared mortality associated with different β-blockers following hospitalization for heart failure between 2001 and 2003. Longitudinal exposure to β-blockers was ascertained from pharmacy databases. Patient characteristics and other medication use were identified from administrative, hospitalization, outpatient, and pharmacy databases. Death was identified from administrative, state mortality, and Social Security Administration databases. Multivariate Cox regression was used to examine the association between different β-blockers and death.
Among 11 326 adults surviving a hospitalization for heart failure, 7976 received β-blockers (atenolol, 38.5%; metoprolol tartrate, 43.2%; carvedilol, 11.6%; and other, 6.7%) during follow-up. The rate (per 100 person-years) of death during the 12 months after discharge varied by exposure and type of β-blocker (atenolol, 20.1; metoprolol tartrate, 22.8; carvedilol, 17.7; and no β-blockers, 37.0). After adjustment for confounders and the propensity to receive carvedilol, the risk of death compared with atenolol was higher for metoprolol tartrate (adjusted hazard ratio [HR], 1.16; 95% confidence interval [CI], 1.01-1.34) and no β-blockers (HR, 1.63; 95% CI, 1.44-1.84) but was not significantly different for carvedilol (HR, 1.16; 95% CI, 0.92-1.44).
Compared with atenolol, the adjusted risks of death were slightly higher with shorter-acting metoprolol tartrate but did not significantly differ for carvedilol in adults with heart failure. Our results should be interpreted cautiously and they suggest the need for randomized trials within real-world settings comparing a broader spectrum of β-blockers for heart failure.
Chronic heart failure (HF) affects more than 5 million Americans and remains the leading cause of hospitalization and a major cause of death among Medicare beneficiaries, with substantial excess medical costs.1 Angiotensin-converting enzyme (ACE) inhibitors,2- 4 aldosterone receptor antagonists,5,6 and angiotensin II receptor blockers7- 10 have beneficial effects on mortality and morbidity in selected patients with HF. Randomized trials have also demonstrated the favorable effect of selected β-adrenergic receptor antagonists (β-blockers) in HF with reduced left ventricular systolic function (eg, extended-release metoprolol succinate,11 bisoprolol,12 carvedilol,13,14 and nebivolol15). Despite generally positive findings for selected β-blockers, arguments against a general class effect come from negative trials of other β-blockers (bucindolol16 and xamoterol17) in advanced HF.
Few head-to-head randomized comparisons of outcomes exist among available β-blockers and have primarily involved carvedilol and shorter-acting metoprolol tartrate.18 To our knowledge, there are no published large-scale evaluations of clinical outcomes comparing available β-blockers, including the widely used atenolol. We previously observed that the adjusted rates of rehospitalization for HF did not vary significantly with atenolol, metoprolol tartrate, or carvedilol use, although information on mortality was unavailable.19 Given that most β-blockers are now generic, it is unlikely that they will be systematically evaluated in future head-to-head randomized trials, yet any outcome differences by β-blocker type could have important clinical implications.
Therefore, we examined the risk of death with different β-blockers among a large, contemporary sample of adults with HF and hypothesized that carvedilol or metoprolol (tartrate or succinate) use would be associated with a lower risk of death compared with atenolol.
Patients were identified from Kaiser Permanente of Northern California, a large integrated health care delivery system in the San Francisco Bay area, and Harvard Pilgrim Health Care, a not-for-profit network-based health care plan operating in Massachusetts, New Hampshire, and Maine. Institutional review boards at collaborating institutions approved the study, and a waiver of informed consent was obtained.
We identified adults hospitalized between January 1, 2001, and December 31, 2003, with a primary discharge diagnosis of HF (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes: 398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 428.0, 428.1, 428.20, 428, 21, 428.22, 428.23, 428.30, 428.31, 428.32, 428.33, 428.40, 428.41, 428.42, 428.43, and 428.9). These codes have a positive predictive value higher than 95% for clinical HF using Framingham criteria.20 We excluded patients if they met any of the following criteria: younger than 18 years, length of stay less than 24 hours, died during index HF hospitalization, and no continuous membership and pharmacy benefit for 12 or more months before the admission date and 12 or more months after discharge or until censoring during the first year after discharge.
Based on the algorithm described in the next paragraph, we characterized β-blocker exposure as the continuation of therapy with prehospitalization β-blockers, switching to a different β-blocker therapy, or the initiation of a new β-blocker therapy at discharge or during follow-up. We searched pharmacy databases during the 12 months before and after the index hospitalization to determine the receipt of any oral β-blocker used in the cohort as identified from generic and brand name as well as National Drug Code and American Hospital Formulary Service codes: acebutolol, atenolol, bisoprolol, carvedilol, labetalol, metoprolol succinate, metoprolol tartrate, nadolol, pindolol, propranolol, sotalol, and timolol. There were no formulary restrictions for β-blockers at participating sites.
We used data on filled prescriptions and estimated timing and duration of β-blocker use based on number of days supplied per prescription and refill patterns.19 For any 2 consecutive prescriptions, we examined the time (in days) between the projected end date of the first prescription and the date of the next filled prescription. Given that dose adjustment is not uncommon, we allowed a “grace period” of 14 days between prescriptions. Thus, if the time between the projected end date of the first prescription and fill date of the next prescription was 14 days or less, we considered that individual to be continually receiving β-blocker therapy. If the refill interval was more than 14 days, then the individual was considered to be not using β-blocker therapy starting the day after the projected end date of the first prescription until the date of next filled prescription, if any. If more than 1 β-blocker prescription was filled on the same day, we used the prescription with the longest estimated day supply. We also accounted for the total number of hospital days during follow-up because hospitalized patients would receive medications from the hospital and not their own outpatient supply. Since concurrent use of multiple β-blockers is not indicated,21 if we observed that a patient filled a prescription for a different β-blocker before the projected end date for an existing β-blocker prescription, the patient was considered to be not using the previous β-blocker as of the fill date of the later prescription for the different β-blocker.
Patients were allowed to switch between β-blockers during the follow-up period, and person-time exposed to each β-blocker of interest was calculated for each individual throughout follow-up. Overall, 1191 patients (14.9% of the cohort) contributed person-time to more than 1 β-blocker during follow-up.
Subjects were followed for up to 12 months after discharge. Deaths were ascertained by probability matching with Social Security Administration Death Master File data through December 31, 2004,22 as well as additional site-specific data sources. For Kaiser Permanente of Northern California, we searched health plan databases (for inpatient deaths and proxy reports of outpatient deaths) and state death files23; for Harvard Pilgrim Health Care, we searched Massachusetts State death files.
Age, gender, time-updated insurance type, calendar year of admission, and length of stay were identified from administrative databases. We used hospital discharge diagnoses or inpatient billing claims to identify hospitalizations for HF occurring during the 12 months before the index hospitalization using the same ICD-9-CM codes previously described. Coexisting illnesses were ascertained from discharge diagnoses and procedures or inpatient claims and ambulatory diagnoses or outpatient physician diagnosis claims during the 12 months before the index hospitalization (a detailed list of ICD-9-CM and Current Procedural Terminology codes and data sources available on request). Prior cardiovascular disease included acute coronary syndromes, other diagnosed coronary disease, percutaneous coronary intervention, coronary bypass surgery, stroke and/or transient ischemic attack, and peripheral arterial disease. We used a previously validated approach24 to identify diabetes mellitus based on a primary discharge diagnosis for diabetes mellitus, 2 or more outpatient diagnoses of diabetes mellitus, and/or 1 or more filled prescriptions for an antidiabetic medication. Hypertension was based on having either 2 or more outpatient diagnoses of hypertension or 1 outpatient diagnosis plus a filled prescription for an antihypertensive medication.25 Dyslipidemia was based on outpatient diagnoses and/or receipt of lipid-lowering therapy. Kidney disease was defined as receipt of renal replacement therapy or a diagnosis of chronic renal failure. We also identified lung disease, liver disease, atrial fibrillation or flutter, known cancer, dementia or psychiatric disorder, and depression.25
Information from pharmacy databases was used to assign prior and postdischarge receipt of relevant cardiovascular medications including ACE inhibitors, angiotensin II receptor blockers, digoxin, diuretics, nitrates, hydralazine, aldosterone receptor antagonists, calcium channel blockers, α-adrenergic receptor antagonists, and lipid-lowering agents.
Statistical analyses were performed using SAS, version 9.1 (Cary, North Carolina). We used the following categories for β-blocker exposure: atenolol, metoprolol tartrate, carvedilol, other β-blockers, and no β-blocker. While use of long-acting metoprolol succinate or bisoprolol was an a priori interest, they were included in the “other β-blockers” category because so few subjects received these agents. Rates of death during the first 12 months after discharge by β-blocker exposure category were calculated using Poisson regression with generalized estimating equations to account for clustering effects within subjects for exposure to specific β-blockers.
As an observational study of clinical practice, concerns exist about treatment selection bias. Our primary goal was to examine differences in mortality with the use of different types of β-blockers, rather than make a treatment vs no treatment comparison, although the latter was also of interest. This approach removes 1 source of selection bias. However, receipt of different β-blockers among β-blocker users may be nonrandom, so we calculated the likelihood of receiving carvedilol (the most specific β-blocker for HF) using the propensity score method.26 The propensity score logistic model included all characteristics in Table 1 and had a C statistic of 0.62.
Multivariate analyses were performed using extended Cox regression with time-dependent covariates. Covariates included variables previously reported to be associated with either receipt of HF therapies or death. Specifically, we adjusted for health plan, age, gender, calendar year of entry, time-varying insurance status, index hospitalization length of stay, prior HF hospitalization, cardiovascular history, other coexisting illnesses, time-varying use of other cardiovascular medications, and propensity score (in deciles) for receiving carvedilol. Because additional models that excluded patients in the highest and lowest propensity score deciles or used greedy pair propensity score matching methods did not change the results (data not shown), we present results only for the overall cohort. We also conducted a subgroup analysis among subjects at 1 site who had documented left ventricular systolic dysfunction (left ventricular ejection fraction <40% or qualitative description of moderate or severe systolic dysfunction).27 Finally, we performed another sensitivity analysis association of new β-blocker use and mortality among the subgroup of patients not receiving β-blockers at admission.
We identified 11 326 patients who survived a hospitalization for HF and met eligibility criteria. Mean age was 73.9 years, and nearly half of the patients were women (Table 1). As expected, there was a high prevalence of prior cardiovascular disease, cardiovascular risk factors, and major comorbidity. Prior to admission, atenolol and metoprolol tartrate were the most frequently used β-blockers (Table 1).
At discharge or during follow-up, 7976 patients received β-blockers. Table 2 gives the characteristics among patients who received atenolol, metoprolol tartrate, carvedilol, or other β-blockers. Compared with patients receiving atenolol (the most frequently prescribed β-blocker), minimal differences were noted for patients receiving metoprolol tartrate, who were slightly younger and more likely to be male; somewhat more likely to have a history of acute coronary syndrome and diagnosed kidney and lung disease (but less likely to have hypertension); and more likely to receive digoxin and aldosterone receptor antagonists. However, compared with atenolol, those receiving carvedilol were younger and more likely to be male and have commercial insurance and lung disease but were less likely to have prior coronary or cerebrovascular disease, hypertension, kidney disease, atrial fibrillation or flutter, dementia or psychiatric disorders, and baseline calcium channel blocker or α-adrenergic receptor antagonist use. However, those receiving carvedilol were more likely to receive ACE inhibitors, angiotensin II receptor blockers, digoxin, diuretics, hydralazine, and aldosterone receptor antagonists at baseline.
The crude rate (per 100 person-years) of death was lowest while receiving carvedilol (17.7), followed by atenolol (20.1), other β-blockers (21.9), and metoprolol tartrate (22.8). Compared with atenolol, observed differences were significant only for metoprolol tartrate (P = .04) and for periods of not using β-blockers (P < .001).
After adjustment for site, calendar year, demographics, insurance status, prior hospitalization for HF, prior cardiovascular history, index hospitalization length of stay, other comorbidity, time-varying use of cardiovascular medications, and propensity to receive carvedilol, receipt of metoprolol tartrate was associated with a 16% (95% confidence interval [CI], 1%-34%) higher relative risk of death compared with atenolol (Table 3). However, compared with atenolol, there was no statistically significant difference in the adjusted risk of death with carvedilol. Of note, there was a higher adjusted relative risk of death during periods of not using β-blockers.
In a subgroup analysis of 2929 patients with documented left ventricular systolic dysfunction, compared with atenolol, we found no significant adjusted differences in mortality with metoprolol tartrate or carvedilol but did observe a significantly higher risk with no β-blocker therapy (Table 3).
Finally, in a sensitivity analysis among 6666 subjects without β-blocker exposure during the 30 days prior to the index hospitalization, 3433 were prescribed β-blockers at discharge or during follow-up. In this subgroup, we found that compared with incident atenolol use, neither incident use of metoprolol tartrate (adjusted hazard ratio [HR] 1.20, 95% CI, 0.99-1.69) or carvedilol (HR, 1.01; 95% CI, 0.68-1.49) was significantly associated with mortality during follow-up. Lack of receipt of any β-blockers was associated with a nearly 2-fold increased risk of death (HR, 1.87; 95% CI, 1.48-2.35).
Within a large cohort of adults hospitalized for HF, we compared the effectiveness of different β-blockers on the subsequent risk of death. A substantial proportion of patients received a β-blocker at discharge and/or during follow-up, with atenolol, shorter-acting metoprolol tartrate, and carvedilol being the most frequently used. Among nearly 8000 high-risk patients with HF receiving β-blockers, there were also notable differences in patient characteristics by type of β-blocker, with carvedilol-treated patients being significantly younger and having a lower comorbidity burden but also receiving more HF-related therapies than those receiving atenolol or metoprolol tartrate.
The high rate of death despite frequent use of various proven therapies in our sample is notable. After adjustment for measured confounders, propensity to receive carvedilol, and longitudinal use of other medications, shorter-acting metoprolol tartrate was still associated with a 16% increased relative risk of death compared with atenolol; there was no significant adjusted difference between carvedilol and atenolol, although there was limited precision for the latter comparison. Results were not materially different in sensitivity analyses in the subset with documented left ventricular systolic function or in analyses restricted only to examining incident use,28 except that the associations for metoprolol tartrate were no longer statistically significant. Of note, not receiving β-blockers was associated with higher adjusted mortality compared with atenolol overall and in subgroup analyses. However, it is important to note that patients contributing to the “no β-blocker group” include those individuals who never were prescribed β-blocker therapy as well as those who received β-blockers at some point during follow-up but had either gaps in treatment or permanently discontinued therapy.
Few large studies exist outside of randomized trial settings that have examined whether outcomes vary by the type of β-blocker in HF. Fonarow and colleagues29 studied 2373 patients hospitalized for new-onset or worsening HF and reduced systolic function in the Organized Program to Initiate Lifesaving Treatment in Hospitalized Patients with Heart Failure (OPTIMIZE-HF) registry and found that most patients treated with β-blockers in this voluntary registry received carvedilol. Compared with not receiving β-blockers at discharge, the adjusted risk of death during the 60 to 90 days after discharge was significantly lower for carvedilol (HR, 0.46; 95% CI, 0.30-0.73) and for metoprolol succinate or bisoprolol (HR, 0.49; 95% CI, 0.28-0.86), but not for the combined group of any other β-blocker (HR, 0.66; 95% CI, 0.37-1.17).29 Analyses comparing the β-blocker categories against each other were not performed, but the direction and similarities of the estimates as well as overlapping confidence limits suggest that no significant differences existed across groups. However, while that study had important strengths, including analyses in the subgroup of patients with reduced systolic function who appeared eligible for β-blocker therapy, it was limited by the modest sample size, low exposure to β-blockers other than carvedilol, short follow-up period, and lack of time-varying information on β-blocker use and other covariates among a selected population. Also, the study was underpowered to examine each of the individual β-blockers. Overall, our study results are generally consistent with those reported in the OPTIMIZE-HF analysis and highlight that patients with HF and reduced systolic function who are not receiving β-blockers tend to experience worse outcomes.
Our study's strengths included the large sample size of patients receiving β-blockers in a contemporary treatment era, geographic diversity, longitudinal data on medications and other confounders, and a multifaceted approach to identify death. Our study had several limitations. In our sample, only atenolol, shorter-acting metoprolol tartrate, and carvedilol were routinely used, while other β-blockers were infrequently used. In addition, information on medication dose taken was unavailable. Most patients receiving β-blockers during follow-up were being treated with β-blockers at admission for the index hospitalization. However, our sensitivity analyses of incident β-blocker use (ie, “new user” study design)28 was consistent with results in the overall cohort. While systematic data were unavailable on systolic function, our results in the subgroup with documented left ventricular systolic function were similar to findings in the larger cohort. We also did not have information on symptom severity, disease-specific quality of life, or functional status.30,31 Data were also not systematically available on selected drugs including over-the-counter aspirin and nonsteroidal anti-inflammatory drugs, lifestyle factors (eg, smoking, alcohol use, diet, and physical activity), and potentially relevant variables (eg, body mass index, blood pressure level, kidney function, and serum electrolyte and lipoprotein levels). Our results also may not be completely generalizable to uninsured populations or other health care settings.
As with any observational study of drug effectiveness, it is vulnerable to residual confounding. To mitigate this, we relied on several different methods. First, our primary comparisons were among patients receiving β-blockers, which removes 1 major source of treatment selection bias. Second, we identified and adjusted for many potential confounding variables, including longitudinal use of other medications. Third, we tested different variations of propensity score method to account for an individual's predicted likelihood of receiving carvedilol.26 Fourth, we conducted several sensitivity analyses that were consistent with the primary findings. However, we note that the observed significant association of higher mortality during periods of not using β-blockers is susceptible to an unmeasured confounder that was associated with a hazard ratio for death of 0.90 or lower or 1.20 or higher.
Our study suggests that additional evidence is needed to clarify whether clinical effectiveness of β-blockers for HF in clinical practice extends beyond the currently approved β-blocker options of carvedilol, metoprolol succinate, and bisoprolol, especially for patients who are receiving a different β-blocker (eg, for treatment of hypertension or coronary heart disease) and then develop clinical HF. Our findings raise the possibility that carvedilol may not have significantly greater effectiveness than atenolol and metoprolol tartrate may be inferior to atenolol for reducing mortality in HF. However, our results should be interpreted cautiously, and future randomized comparisons should be considered that include a broader set of different β-blockers in this population. If our observations are confirmed, it could have important implications for optimizing the treatment and outcomes in high-risk patients with HF.
Correspondence: Alan S. Go, MD, Division of Research, Kaiser Permanente of Northern California, 2000 Broadway St, Third Floor, Oakland, CA 94612-2304 (Alan.S.Go@kp.org).
Accepted for Publication: May 13, 2008.
Author Contributions:Study concept and design: Go, Yang, Gurwitz, and Platt. Acquisition of data: Go, Yang, Lane, and Platt. Analysis and interpretation of data: Go, Yang, Gurwitz, Hsu, and Platt. Drafting of the manuscript: Go and Yang. Critical revision of the manuscript for important intellectual content: Go, Yang, Gurwitz, Hsu, Lane, and Platt. Statistical analysis: Yang. Obtained funding: Go, Lane, and Platt. Administrative, technical, and material support: Go, Yang, Lane, and Platt. Study supervision: Go.
Financial Disclosure: None reported.
Funding/Support: This project was funded under contract HHSA290-2005-0033-I from the Agency for Healthcare Research and Quality, US Department of Health and Human Services as part of the Developing Evidence to Inform Decisions about Effectiveness (DEcIDE) program.
Disclaimer: The authors of this report are responsible for its content and statements within should not be construed as being endorsed by the Agency for Healthcare Research and Quality or the US Department of Health and Human Services.
Additional Contributions: Jim Livingston, MBA, Inna Dashevsky, and Ning Hernandez provided expert technical assistance.