LVEF indicates left ventricular ejection fraction.
aAdjusted for propensity score categorized in deciles.
NYHA indicates New York Heart Association.
eTable 1. ICD-10 Codes for HF Diagnosis
eTable 2. Variables Included in Propensity Score, With Sources of Data and ICD-10 and ATC Codes
eTable 3. Hazard Ratios With 95% CIs From Full Multivariate Model of All-Cause Mortality Adjusted for Propensity Score From 10 Imputed Datasets
eTable 4. All-Cause Mortality Among Patients With Heart Failure and Reduced Ejection Fraction and Using Carvedilol or Metoprolol Succinate: Follow-Up Truncated at 1 Year
eTable 5. Secondary Outcome of Cardiovascular Mortality Comparing Patients With Heart Failure and Reduced Ejection Fraction and Using Carvedilol or Metoprolol Succinate
eTable 6. Baseline Characteristics of 1:1 Propensity Score-Matched Cohort of Patients With Heart Failure and Reduced Ejection Fraction Included in Robustness Analysis
Pasternak B, Svanström H, Melbye M, Hviid A. Association of Treatment With Carvedilol vs Metoprolol Succinate and Mortality in Patients With Heart Failure. JAMA Intern Med. 2014;174(10):1597-1604. doi:10.1001/jamainternmed.2014.3258
The β-blockers carvedilol and metoprolol succinate both reduce mortality in patients with heart failure (HF), but the comparative clinical effectiveness of these drugs is unknown.
To investigate whether carvedilol is associated with improved survival compared with metoprolol succinate.
Design, Setting, and Participants
Cohort study of patients with incident HF with reduced left ventricular ejection fraction (LVEF) (≤40%) who received carvedilol (n = 6026) or metoprolol succinate (n = 5638) using data from a Danish national HF registry linked with health care and administrative databases.
Main Outcomes and Measures
All-cause mortality (primary outcome) and cardiovascular mortality (secondary outcome) were analyzed using Cox regression with adjustment for a propensity score, derived from a range of clinical, socioeconomic, and demographic characteristics.
The mean (SD) age of the patients was 69.3 (9.1) years, 71% were men, and 51% were hospitalized at index HF diagnosis. During a median (interquartile range) 2.4 (1.0-3.0) years of follow-up, 875 carvedilol users and 754 metoprolol users died; the cumulative incidence of mortality was 18.3% and 18.8%, respectively. The adjusted hazard ratio for carvedilol users vs metoprolol users was 0.99 (95% CI, 0.88 to 1.11), corresponding to an absolute risk difference of –0.07 (95% CI, –0.84 to 0.77) deaths per 100 person-years. Estimates were consistent across subgroup analyses by sex, age, levels of LVEF, New York Heart Association classification, and history of ischemic heart disease. A higher proportion of carvedilol users achieved the recommended daily target dose (50 mg; 3124 [52%]) than did metoprolol users (200 mg; 689 [12%]); among patients who reached the target dose, the adjusted hazard ratio was 0.97 (95% CI, 0.72-1.30). A robustness analysis with 1:1 propensity score matching confirmed the primary findings (hazard ratio, 0.97 [95% CI, 0.84-1.13]). The adjusted hazard ratio for cardiovascular mortality was 1.05 (95% CI, 0.88-1.26).
Conclusions and Relevance
These findings from real-world clinical practice indicate that the effectiveness of carvedilol and metoprolol succinate in patients with HF is similar.
β-Blockers represent the mainstay of treatment for heart failure (HF), with placebo-controlled randomized trial evidence of prolonged survival with use of metoprolol succinate (controlled release/extended-release metoprolol formula), carvedilol, and bisoprolol fumarate.1- 3 Among the 3 β-blockers with trial evidence of efficacy in HF, no individual drug is recommended over the other by major guidelines,4,5 thus implying similar efficacy. However, whereas the magnitude of mortality reduction vs placebo was similar in key trials,1- 3 to our knowledge, there are no published trials with mortality end points that compare any 2 of the 3 specific agents head-to-head.
Metoprolol and bisoprolol are β1 receptor selective, whereas carvedilol has β1, β2, and α1 receptor–blocking properties. Furthermore, carvedilol exhibits antioxidant activity and is an inhibitor of endothelin.6,7 These pharmacological differences have been suggested to underlie a potential benefit of carvedilol over other β-blockers in patients with HF. This hypothesis was tested in the Carvedilol or Metoprolol European Trial (COMET), which found a significant 17% survival benefit with use of carvedilol over metoprolol.8 However, COMET studied a short-acting immediate-release metoprolol formulation (metoprolol tartrate) that is not approved for HF and the appropriateness of the target metoprolol dose used in the trial has been the subject of extensive debate.9- 12 A recent network meta-analysis of randomized trials did indirect head-to-head comparisons between β-blockers and reported an odds ratio consistent with a potential benefit in favor of carvedilol over metoprolol (0.80 [95% CI, 0.59-1.08]),13 but a large proportion of metoprolol-treated patients were derived from COMET, which made interpretation relevant to current clinical practice difficult.
To investigate whether carvedilol use is associated with improved survival compared with metoprolol succinate use, we conducted an observational study of real-world patients with HF in Denmark.
We conducted a national cohort study in Denmark of adults 50 to 84 years of age who had an incident diagnosis of HF with reduced left ventricular ejection fraction (LVEF) and received treatment with carvedilol or metoprolol succinate. In head-to-head analyses with follow-up of up to 3 years, we investigated all-cause mortality (primary outcome) and cardiovascular mortality (secondary outcome), adjusting for potential confounders through propensity score methods. The study was approved by the Danish Data Protection Agency and the Danish HF Registry. Ethics approval and informed consent are not required for register-based research in Denmark.
Patients with incident HF were identified from the Danish HF Registry,14 a database established in 2003 with the purpose of documenting the quality of HF care in Denmark. All hospital departments and clinics in the country that care for inpatients and outpatients with HF are invited to participate and are recommended to include all patients with a clinical diagnosis of HF consistent with the criteria of the European Cardiology Society. Included in the registry are patients with a first-time primary diagnosis of HF. Additional details and a description of all other databases used for this study, which also included the Central Person Register,15 Statistics Denmark, the National Patient Register,16 the National Prescription Registry,17 and the Cause of Death Register,18 are provided in the eMethods in the Supplement.
Eligible for inclusion were all patients 50 to 84 years of age identifiable in the HF Registry between January 1, 2003, and December 31, 2012, who had reduced LVEF (≤40%). The study cohort included every eligible patient who had filled at least 2 prescriptions for a study medication within a maximum of 180 days following diagnosis (as a minimum standard for adherence; requirement of first prescription fill within 60 days following diagnosis and second prescription fill within 120 days of the first), did not receive multiple β-blockers on the same date, and had been residing in Denmark for at least 2 years.
To account for potential confounders, we used propensity score methods.19 By means of logistic regression analysis, the individual probability of carvedilol treatment was estimated given a total of 126 variables including demographic variables, socioeconomic characteristics, comorbidities, medications, HF characteristics, lifestyle factors, measures of health care use, and a number of 2-way interactions (eTable 2 in the Supplement). Multiple imputation (Markov chain Monte Carlo method) was used to handle missing data20; all analyses were conducted using 10 imputed data sets.
Patients were observed from the date of the second prescription for a study β-blocker and up to 3 years; censoring criteria were switch to any other β-blocker than the first, loss to follow-up (emigration, disappearance), and end of the study period (June 30, 2013). In the analysis of cardiovascular mortality, other-cause death was an additional censoring criterion; data on causes of death were available through 2011, and the cohort was truncated accordingly for this analysis.
Cumulative incidence curves were generated using the Kaplan-Meier method. Cox proportional hazards regression analysis was used to estimate hazard ratios with 95% confidence intervals; time since study entry was the timescale. Models were adjusted for propensity scores categorized in deciles. The proportional hazards assumption was assessed by a Wald test for interaction between treatment status and time and was fulfilled in all primary and secondary outcome analyses. The adjusted absolute risk difference per 100 person-years of carvedilol use was estimated as (adjusted hazard ratio [aHR] – 1) × crude rate among metoprolol users. Subgroup analyses were conducted by sex, age, New York Heart Association (NYHA) class, LVEF, and history of ischemic heart disease. Differences were considered statistically significant when the 95% confidence interval did not overlap 1.0. Analyses were performed using SAS software, version 9.3.
We performed preplanned robustness analyses, as follows. (1) Our primary analysis included both patients using study β-blockers before HF diagnosis and those in whom this treatment was initiated after diagnosis. Prevalent users may have survived important adverse events of medication and might be at lower risk for the outcome studied. Therefore, we implemented the new-user design,21 defining new users as patients who initiated treatment after HF diagnosis and had no study drug use in the previous 2 years. (2) Given the possible, albeit not definitive, importance of β-blocker dose,22- 25 we did an analysis restricted to patients who achieved the recommended target daily dose (50 mg for carvedilol and 200 mg for metoprolol).4,5 For this analysis, the dose was estimated from the tablet strength and patients were observed from the day they reached the target dose. (3) As an alternative method of confounder control, we adjusted for an empirical risk score for mortality (described in the eMethods in the Supplement).26 (4) We used an alternative time-updated definition of drug exposure, in which each prescription generated 3 months of drug use, and compared mortality while on treatment with carvedilol or metoprolol. Events that occurred during any treatment pauses or after treatment discontinuation were defined as occurring off treatment and did not contribute to the analysis. (5) On the assumption that propensity score matching might provide superior confounder control than adjustment or at least reduce any heterogeneity, we frequency-matched carvedilol and metoprolol users (1:1 ratio) according to propensity score using the greedy 5→1 digit-matching algorithm.27,28
A total of 6026 carvedilol users and 5638 metoprolol users were included (Figure 1). Mean (SD) age in the cohort was 69.3 (9.1) years, 71% were men, 51% had been hospitalized at index diagnosis, and most patients were in NYHA class II or III; the distribution of most baseline characteristics was similar between the 2 groups (Table 1). History of cardiomyopathy was more common among carvedilol users, as was use of loop diuretics and aldosterone antagonists; a higher proportion of carvedilol users were in the lowest LVEF category (<25%) and the higher NYHA categories. History of acute coronary syndrome, other ischemic heart disease, percutaneous coronary intervention or coronary artery bypass grafting, arrhythmia, and cardiovascular hospitalization or emergency department visit was more common among metoprolol users, as was use of digoxin, nitrates, platelet inhibitors, anticoagulants, and lipid-lowering drugs. There were missing values in 6 of the variables (Table 1); the relative efficiency of the imputation procedure was greater than 99%. The C statistic for the propensity score model was 0.80 in each imputation set.
The median (interquartile range) overall follow-up was 2.4 (1.0-3.0) years, 2.7 (1.1-3.0) years in carvedilol users and 2.1 (0.9-3.0) years in metoprolol users. The median (interquartile range) time on treatment, defined as the time from start of follow-up through the last prescription, was 2.3 (0.9-2.9) years in carvedilol users and 1.8 (0.6-2.9) years in metoprolol users. A total of 18 patients in each group were censored as a result of switching to another β-blocker (n = 16 among carvedilol users; n = 14 among metoprolol users) and loss to follow-up (n = 2, carvedilol; n = 4, metoprolol).
Among carvedilol users, 875 deaths occurred, whereas there were 754 deaths among metoprolol users, corresponding to a cumulative incidence of the primary outcome of all-cause mortality of 18.3% and 18.8%, respectively (Figure 2). After adjustment for propensity score, the risk of mortality did not differ significantly between carvedilol and metoprolol users (aHR, 0.99 [95% CI, 0.88-1.11]) (Figure 2; the full multivariable model is shown in eTable 3 in the Supplement). The corresponding adjusted absolute risk difference was −0.07 (95% CI, −0.84 to 0.77) deaths per 100 person-years. Results were similar with follow-up truncated at 1 year (aHR, 0.98 [95% CI, 0.83-1.16]) (eTable 4 in the Supplement). The risk of the secondary outcome of cardiovascular mortality was not significantly different between carvedilol and metoprolol users (aHR, 1.05 [95% CI, 0.88-1.26]) (eTable 5 in the Supplement).
In subgroup analyses (Figure 3), the aHRs for carvedilol vs metoprolol users were similar in subgroups stratified by sex, age, LVEF, NYHA class, and history of ischemic heart disease.
First, in the analysis restricted to new users of the study β-blockers, the aHR for mortality was 0.95 (95% CI, 0.81-1.12) (Table 2). Second, a higher proportion of carvedilol users achieved the recommended daily target dose (50 mg; 3124 [52%]) than did metoprolol users (200 mg; 689 [12%]). Among these patients, the risk of mortality was not significantly different between carvedilol and metoprolol users (aHR, 0.97 [95% CI, 0.72-1.30]) (Table 2). Third, we estimated a risk score for mortality and included this as an adjustment variable (instead of propensity score). The C statistic for the risk score model was between 0.91 and 0.92 (numbers varied with each imputation set). Adjustment for risk score yielded an estimate that was very similar to that in the primary analysis (aHR, 0.99 [95% CI, 0.89-1.09]) (Table 2). Fourth, the aHR in the analysis using an on-treatment definition of β-blocker exposure was 0.96 (95% CI, 0.84-1.08) (Table 2). Fifth and finally, 1:1 propensity score matching yielded a cohort that was well balanced on all baseline characteristics (eTable 6 in the Supplement). The hazard ratios for all-cause and cardiovascular mortality in the propensity score–matched cohort were 0.97 (95% CI, 0.84-1.13) and 1.03 (95% CI, 0.82-1.29), respectively.
In this large contemporary national cohort study of patients with HF with reduced LVEF, we found no significant difference in all-cause mortality between carvedilol and metoprolol succinate users. The primary findings were consistent through various analyses, including the secondary outcome of cardiovascular mortality, key subgroups, and robustness analyses. Given the limits of the confidence intervals, this study could rule out a relative difference in mortality of more than 12% and an absolute difference of 1 death or more per 100 patient-years. This suggests that any difference between carvedilol and metoprolol succinate, if it exists, is unlikely to be clinically meaningful.
We designed this study to investigate whether carvedilol is associated with improved survival compared with metoprolol succinate, a hypothesis based on the different adrenergic receptor profiles of the 2 drugs, other mechanistic differences, and results of the COMET trial and a recent network meta-analysis.6- 8,13 The magnitude of reduction in mortality vs placebo was similar in the key carvedilol and metoprolol succinate trials,1,3 but the 2 drugs have not been compared in a head-to-head trial adequately powered to investigate mortality. Given the required size and associated cost, it is unlikely that a randomized trial that resolves this issue will be conducted in the near future. Therefore, we did a cohort study of real-world patients in clinical practice. Our results are consistent with the notion that no clinically relevant differences between carvedilol and metoprolol succinate exist. To our knowledge, no directly comparable studies have been published. A few observational comparative effectiveness studies have been conducted, but these only included metoprolol tartrate and did not have data on LVEF and NYHA class29,30 or were conducted at a single center with a limited number of adjustment variables.31
Given the paucity of published head-to-head data, our study expands on the available evidence of the comparative effectiveness of carvedilol and metoprolol succinate therapy in HF by contributing an adequately powered analysis of mortality in a well-characterized study cohort. It lends support to current treatment recommendations,4,5 which do not explicitly support the use of one β-blocker with proven mortality benefit in HF over the other and thereby regard the effectiveness of these drugs as equivalent.
Our study has strengths and limitations. Treatment was not randomly assigned and could therefore have been influenced by unmeasured factors that, if also associated with the outcome, could have introduced confounding. To reduce this possibility, we took into account a range of potential confounders through the use of a comprehensive propensity score model. Of note, although some baseline differences between the treatment groups were present, such as higher prevalence of ischemic heart disease among metoprolol users, potentially introducing a certain level of heterogeneity that might not have been captured by adjustment for propensity score, the main results were replicated in a 1:1 propensity score–matched analysis, which was based on a subcohort that was well balanced on baseline characteristics. Despite this, our findings should be interpreted considering the observational design of the study and need replication in an independent population. Furthermore, a study of clinical effectiveness in real-world patients should be seen as complementary rather than comparable to a trial.
The use of total population registers allowed detailed characterization of included patients, minimal loss to follow-up, and exact assessment of outcomes throughout the study period. Concerns may be raised that a higher proportion of carvedilol users achieved the target dose than did metoprolol users; assuming dose-dependent effects, this could mask a true benefit of metoprolol over carvedilol. However, these concerns are not supported by the analysis restricted to patients who reached the target dose, which was in line with the primary analysis. Furthermore, the issue of dose-dependent effects is not settled; although some data support this notion,22,23 a meta-analysis found that the magnitude of heart rate reduction was associated with reduction in mortality, whereas β-blocker dose was not.25
The findings are likely generalizable to other populations with similar characteristics. It should be kept in mind, however, that included patients represent a subset (ie, those starting β-blocker treatment) of those included in an HF registry. Therefore, frail patients with short expected survival may not have had treatment initiated and would not have been included. Furthermore, by design, our study did not include those who died during index hospitalization and those with poor adherence. Indeed, cumulative mortality reached approximately 18% during the median 2.4-year follow-up, which is lower than reported in other observational studies (these, on the other hand, included a large proportion of patients not receiving evidence-based HF treatment).32
This large cohort study of real-world patients with HF with reduced LVEF contributes new data on the comparative effectiveness of carvedilol vs metoprolol succinate and supports similar clinical effectiveness of these 2 evidence-based drugs.
Accepted for Publication: May 9, 2014.
Corresponding Author: Björn Pasternak, MD, PhD, Department of Epidemiology Research, Statens Serum Institut, Artillerivej 5, 2300 Copenhagen S, Denmark (firstname.lastname@example.org).
Published Online: August 31, 2014. doi:10.1001/jamainternmed.2014.3258.
Author Contributions: Dr Pasternak 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 concept and design: All authors.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Pasternak.
Critical revision of the manuscript for important intellectual content: Svanström, Melbye, Hviid.
Statistical analysis: Pasternak, Svanström.
Administrative, technical, or material support: Melbye.
Study supervision: Melbye, Hviid.
Conflict of Interest Disclosures: None reported.
Previous Presentation: This study was presented at the European Society of Cardiology Congress; August 31, 2014; Barcelona, Spain.
Additional Contributions: We thank the Danish HF Registry for providing data for this project.
Correction: This article was corrected online September 10, 2014, for missing previous presentation information.