Figure. Selection of patients for analyses. CVD indicates cardiovascular disease.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Parker ED, Margolis KL, Trower NK, Magid DJ, Tavel HM, Shetterly SM, Ho PM, Swain BE, O’Connor PJ. Comparative Effectiveness of 2 β-Blockers in Hypertensive Patients. Arch Intern Med. 2012;172(18):1406–1412. doi:10.1001/archinternmed.2012.4276
Background Randomized controlled trials have demonstrated the efficacy of selected β-blockers for preventing cardiovascular (CV) events in patients following myocardial infarction (MI) or with heart failure (HF). However, the effectiveness of β-blockers for preventing CV events in patients with hypertension has been questioned recently, but it is unclear whether this is a class effect.
Methods Using electronic medical record and health plan data from the Cardiovascular Research Network Hypertension Registry, we compared incident MI, HF, and stroke in patients who were new β-blocker users between 2000 and 2009. Patients had no history of CV disease and had not previously filled a prescription for a β-blocker. Cox proportional hazards regression was used to examine the associations of atenolol and metoprolol tartrate with incident CV events using both standard covariate adjustment (n = 120 978) and propensity score–matching methods (n = 22 352).
Results During follow-up (median, 5.2 years), there were 3517 incident MI, 3272 incident HF, and 3664 incident stroke events. Hazard ratios for MI, HF, and stroke in metoprolol tartrate users were 0.99 (95% CI, 0.97-1.02), 0.99 (95% CI, 0.96-1.01), and 0.99 (95% CI, 0.97-1.02), respectively. An alternative approach using propensity score matching yielded similar results in 11 176 new metoprolol tartrate users, who were similar to 11 176 new atenolol users with regard to demographic and clinical characteristics.
Conclusions There were no statistically significant differences in incident CV events between atenolol and metoprolol tartrate users with hypertension. Large registries similar to the one used in this analysis may be useful for addressing comparative effectiveness questions that are unlikely to be resolved by randomized trials.
In the treatment of hypertension, β-blockers are widely used and are one of the drug classes recommended as initial treatment in hypertension guidelines based on reduction of morbidity and mortality in placebo-controlled trials.1-5 However, following the publication of 2 large trials that found that atenolol-based regimens were less effective than other antihypertensive drugs for prevention of cardiovascular (CV) events in patients with hypertension,6,7 the first-line status of β-blockers has increasingly been called into question.3,8-11 A recent meta-analysis including these studies found that β-blockers were inferior to other agents primarily with regard to stroke prevention, but the authors and editorialist pointed out that data on β-blockers other than atenolol were sparse enough that it is unclear whether this conclusion applies to the entire β-blocker class.9,12 A second meta-analysis and editorial echoed these findings and concerns.11,13
Within the drug class of β-blockers, there are differences in pharmacokinetic properties.14,15 Differences in lipophilicity, bioavailability, and metabolism between atenolol and metoprolol tartrate may have relevance for protecting the heart.10,11Despite these differences, it is unlikely that they will be compared head to head in a randomized controlled trial. Therefore, we sought to compare the effectiveness of 2 commonly used β-blockers, using data from a hypertension registry from 3 large integrated health care delivery systems. We compared the incidence of myocardial infarction (MI), stroke, and heart failure (HF) in adult hypertensive patients who were new users of atenolol and metoprolol tartrate.
This report is derived from the Hypertension Registry of the Cardiovascular Research Network (CVRN). The registry includes all adult patients identified as having hypertension between 2000 and 2009 at 3 large integrated health care delivery systems: HealthPartners of Minnesota, Kaiser Permanente Colorado, and Kaiser Permanente Northern California. Electronic data on longitudinal blood pressure (BP) measurements, prescription drugs, laboratory test results, diagnoses, and health care utilization were available from electronic health records and administrative databases at all sites. Data from each of the health plans were restructured into a common, standardized format with identical variable names, definitions, labels, and coding.
We defined hypertension using criteria adapted from previous CVRN studies16-20 based on outpatient BP readings, diagnostic codes from outpatient and hospital records, pharmacy prescriptions, and laboratory results. Patients entered the registry on the date they first met 1 (or more) of the following criteria: (1) 2 consecutive elevated BP measurements (ie, systolic BP [SBP] ≥140 mm Hg and/or diastolic BP [DBP] ≥90 mm Hg, or ≥130/80 mm Hg in the presence of diabetes mellitus or chronic kidney disease [CKD]); (2) 2 diagnostic codes for hypertension (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 401.x-405.x) recorded on separate dates; (3) 1 diagnostic code for hypertension plus prescription for an antihypertensive medication; or (4) 1 elevated BP measurement plus 1 diagnostic code for hypertension. Blood pressure readings from emergency and urgent care settings were excluded because they were found to be consistently higher than other ambulatory measurements in the same patients in similar periods. To confirm that the algorithms designed to identify hypertensive patients were valid and that the analytic data accurately reflected the source data, we conducted a review of 450 randomly selected medical charts (150 from each site). We confirmed that hypertension was in fact incident on the date assigned by the algorithm in 96% of cases, and agreement on BP values between the electronic database and medical chart records was 98%.
Patient age and sex were available for all patients from membership databases. Race/ethnicity was obtained from outpatient registration data, hospital discharge records, member satisfaction surveys, and other research survey data sets and was available for 85% of cohort members. Systolic BPs and DBPs measured within 2 months prior to the initiation of a β-blocker therapy and approximately 6 months (±60 days) after the initiation of a β-blocker therapy were included. Pharmacy records were used to identify dates of treatment with β-blockers and other antihypertensive drug classes used within 90 days of starting the β-blocker therapy.
Cardiovascular disease (CVD) was identified using diagnoses and procedure codes from inpatient and ambulatory records. These included ischemic heart disease (ICD-9-CM diagnosis codes 410.x-414.xx); stroke (ICD-9-CM diagnosis codes 430.xx-434.xx, 436.xx, 852.0, 852.2. 852.4, and 853.0); peripheral vascular disease (ICD-9-CM diagnosis codes 441.3-441.7, 443.9, 444.0, and 444.2); and congestive HF (ICD-9-CM diagnosis codes 428.xx, 402.xx, and 398.91). Incident MI (ICD-9-CM code 410.xx), HF, and stroke events were defined using the primary International Classification of Diseases, Ninth Revision (ICD-9) codes from a discharge from an inpatient stay.
Other comorbidities included in the analysis were diabetes mellitus, CKD, and lipid disorders. Diabetes was defined by (1) 2 outpatient diagnoses or 1 primary inpatient discharge diagnosis of diabetes mellitus (ICD-9-CM code 250.x); (2) a prescription for any antidiabetic medication other than metformin or thiazolidinediones; (3) a prescription for metformin or a thiazolidinedione plus a diagnosis of diabetes; or (4) a hemoglobin A1c value higher than 7% or 2 fasting plasma glucose values of 126 mg/dL or higher (to convert to millimoles per liter, multiply by 0.0555) on separate dates. Chronic kidney disease was defined by (1) 2 consecutive serum creatinine values that yield estimated glomerular filtration rates lower than 60 mL/min or (2) an International Classification of Diseases, Ninth Revision (ICD-9) diagnostic code for CKD (ICD-9-CM codes 585.1-585.9). Lipid disorders were identified by ICD-9-CM codes 272.x.
We used a new user design, which restricts the analysis to persons under observation at the start of the current course of treatment.21 The study population included all patients 18 years or older with hypertension during 2000 through 2009, who were started on therapy with either atenolol or metoprolol tartrate after the date of first diagnosis with no prior use of any β-blocker for at least 12 months (n = 193 123). Previous use of any other class of antihypertensive drug was not an exclusion. Prescription databases were searched as far back as 1996 or to health plan enrollment if that occurred after 1996. Other β-blockers, including metoprolol succinate, were not used frequently enough during the years of the study to be included in the analysis. We excluded pregnant women (n = 346). In addition, we excluded 46 809 patients who had evidence of CVD before starting therapy with atenolol or metoprolol tartrate. These exclusions were based on the previously described CV diagnosis codes as well as procedure codes for cardiac bypass surgery (Current Procedural Terminology [CPT] codes 33510-33523 and 33533-33536) and percutaneous coronary interventions (CPT codes 92980-92996). To exclude patients with suspected CVD, we also excluded 24 990 patients with a visit to cardiology specialist within the year prior to starting the β-blocker therapy, leaving 120 978 patients for this analysis (Figure).
All statistical analyses were completed using SAS version 9.2 (SAS Institute Inc). Baseline characteristics were compared between patients started on atenolol therapy vs metoprolol tartrate therapy using means and standard deviations for continuous variables and percentages for categorical and binary variables. Cox proportional hazards models were used to compare time with outcome events between atenolol and metoprolol tartrate. Follow-up time was computed in days from the day following the first dispensing of the new β-blocker to the date of the first observed outcome event, termination of enrollment, or December 31, 2009, whichever occurred first. Patients who were lost to follow-up were censored at the last point of contact. Multivariable models were adjusted for year of β-blocker therapy initiation, age, sex, number of visits in the prior year, SBP at the start of β-blocker therapy, lipid disorder, diabetes mellitus, CKD, and use of other antihypertensive medications. In a supplemental analysis of 68 882 patients in whom we had follow-up BP data, we used linear regression to examine the effect of atenolol and metoprolol tartrate on lowering SBP and DBP 6 months after the start of the β-blocker therapy.
Because this is an observational study and patients were not randomized to receive either treatment, we also used alternative strategies to minimize confounding by indication. To minimize confounding by indication, we also ran a conditional logistic regression matched on propensity score. A logistic model (which included all the variables in Table 1 except for DBP) was used to generate a propensity score for the probability of being prescribed metoprolol tartrate. We then used a 5-digit greedy 1:1 matching algorithm23 to match metoprolol tartrate users to atenolol users based on propensity score. After conducting the propensity matching, there were 99 626 unmatched patients, leaving 22 352 matched patients for statistical analyses of adverse CV events and 13 908 matched patients with 6-month follow-up BPs for the analyses of BP lowering. The selection of patients for the analyses is shown in the Figure. A second alternative strategy was to conduct a sensitivity analysis excluding patients who had events in the first 12 months of follow-up so as to exclude those with CVD not excluded by diagnosis codes or visits to a cardiologist that could have had an impact on prescribing behavior.
The baseline characteristics for this cohort of new β-blocker users are given in Table 1. A total of 120 978 patients without history of CVD events from the CVRN Hypertension Registry initiated treatment with either atenolol or metoprolol tartrate between 2000 and 2009. During this period atenolol was used in approximately 10-fold more patients than metoprolol tartrate. Patients who filled a prescription for metoprolol tartrate tended to be older, have a government insurance payer, and have more ambulatory visits. Metoprolol tartrate users had slightly lower SBPs and DBPs at the start of β-blocker treatment, were more likely to be using other antihypertensive medications, and more often had lipid disorders, diabetes, and CKD.
During the follow-up period (median, 5.2 years), there were 3517 incident MIs, 3272 incident HF hospitalizations, and 3664 incident strokes. Multivariable Cox proportional hazards regression yielded hazard ratios of 0.99, 0.99, 0.99, and 0.98 and narrow 95% confidence intervals that included the null value for MI, HF, stroke, and any CV event, respectively (Table 2). In the propensity score–matched Cox proportional hazards models, the hazard ratios for MI, HF, stroke, and any CV event were virtually identical to the multivariable results with narrow 95% confidence intervals that included the null value (Table 2). In sensitivity analyses excluding patients who had events in the first 12 months of follow-up, the hazard ratios were virtually unchanged (data not shown).
Estimates and standard errors of the supplemental analysis of the BP-lowering effects of the 2 β-blockers in the subgroup with follow-up measures are given in Table 3. In multivariable analysis of new β-blocker users, at baseline there were statistically significant differences between atenolol and metoprolol tartrate users in SBPs (148.5 and 145.4 mm Hg, respectively; P < .001) and DBPs (84.2 and 82.5, respectively; P < .001). At the 6-month follow-up, SBPs were 137.4 and 137.5 mm Hg in the atenolol- and metoprolol tartrate-treated patients, respectively (P = .82). At 6 months, DBPs were 77.3 and 77.7 mm Hg in the atenolol- and metoprolol tartrate–treated patients, respectively (P = .005). There was no statistically significant difference in change in SBP and a small but statistically significant difference in change in DBP (5.9 and 5.5 mm Hg for atenolol and metoprolol tartrate, respectively P = .005). The propensity score–matched analysis of BP lowering had similar results when comparing new atenolol and metoprolol tartrate users in SBP (144.2 and 143.3 mm Hg, respectively; P = .007) and DBP (81.3 and 80.2 mm Hg, respectively; P < .001). At the 6-month follow-up, there were no statistically significant differences between atenolol and metoprolol tartrate users in SBP or DBP. In the propensity-matched model, the mean BP lowering was slightly greater in atenolol vs metoprolol tartrate users (7.7 and 6.7 mm Hg, respectively; P = .02). Atenolol lowered DBP slightly more than metoprolol tartrate (4.7 and 3.4 mm Hg, respectively; P < .001).
The objective of this study was to assess the comparative effectiveness of 2 β-blockers, atenolol and metoprolol tartrate, in patients without a history of CVD. To our knowledge, this study is among the first to address this important clinical question. In this retrospective cohort study comparing patients initiating β-blocker treatment with either atenolol or metoprolol tartrate, there were no statistically significant differences in rates of incident MI, HF, or stroke after adjusting for potential confounders. In addition, there were no statistically significant differences in SBP-lowering effects comparing atenolol and metoprolol tartrate.
Until recently, β-blockers had been widely recommended as first-line therapy for hypertension,1-5 but many of the trials supporting their use had given investigatorsthe choice of using either a thiazide diuretic or β-blocker alone or in combination as “conventional therapy.” The combination compared favorably against other antihypertensive drugs classes for prevention of CV events.1,24 The use of β-blockers as a first-line therapy has recently been challenged based on evidence of a weak effect on stroke25 and the absence of an effect on coronary heart disease25-27 compared with placebo, as well as inferiority compared with other treatments for total mortality, coronary heart disease, and stroke.6,7,28 Meta-analyses and a Cochrane review of recent trials that looked specifically at β-blockers used as monotherapy or as the first-line drug in a stepped care approach concluded that the evidence did not support use of β-blockers as a first-line therapy.11 Based on these findings, recently issued guidelines have relegated β-blockers to third- or fourth-line treatment for uncomplicated hypertension.29
β-blockers differ in selectivity for the β1- and β2- and α-adrenergic receptors, lipophilicity, penetration across the blood-brain barrier, duration of action, vasodilation properties, and type 3 antiarrhythmic activity.14,15,30 Different types of β-blockers may be indicated depending on patient profiles and tolerances. Given that most of the evidence comes from trials where atenolol was the β-blocker used,11 it is unclear if the observed effects of β-blockers in comparison with other antihypertensive medications are due to properties of atenolol or the entire class of β-blockers. However, there have been no trials comparing the different subtypes of β-blockers. While both atenolol and metoprolol tartrate are both β1-adrenergic receptors, they differ in lipophilicity, bioavailability, and metabolism.10,11,31 Metoprolol is lipid soluble and tends to have highly variable bioavailability and a short plasma half-life. In contrast, atenolol is more water soluble, shows less variance in bioavailability, and has a longer plasma half-life. Despite these differences, both drugs have the effect of increasing vagal tone and causing a reduction in sympathetic outflow, likely via peripheral β-adrenergic blockade.32,33 Our findings that there are no differences between atenolol and metoprolol tartrate in event rates and effectiveness at BP lowering in a cohort of adults without prior CV events suggest that the unfavorable trial data with atenolol may also apply to other β-blockers.
As with any observational study, there are potential limitations and caveats. We were unable to compare atenolol with any β-blocker other than metoprolol tartrate because of the low use of other agents in our study population during the years of observation. The use of metoprolol succinate, a once-daily drug that may have better adherence rates compared with twice-daily metoprolol tartrate, has been increasing owing to the availability of generic versions in recent years, but the shift away from β-blockers after 2007 may make comparative effectiveness analyses more difficult.
Most importantly, patients were not randomly assigned to treatment with either atenolol or metoprolol tartrate. The decision on the part of the clinician to choose one drug over another may be related to patient characteristics associated with BP control or CV risk or physician characteristics associated with differences in quality of care. To reduce the potential bias related to confounding by indication, we took 2 approaches: (1) we used a new user design21,34 and restricted the sample to those patients with no evidence of diagnosed or suspected CVD34 and (2) we used propensity score matching to ensure that patients were comparable with regard to baseline covariates and the probability of receiving each treatment.
Despite these robust methods, no observational study can rule out the impact of unmeasured confounding. If unmeasured variables associated with poorer prognosis were more common in patients prescribed metoprolol tartrate, it could mask a beneficial effect of metoprolol. We excluded patients who had seen a cardiologist in the 12 months prior to the initiation of β-blocker therapy, but this strategy may have been insufficient to rule out suspected CVD. However, in a recent study using data from one of the study sites, we found no evidence of suspected heart disease in audits of physician medical chart notes in 240 patients lacking specific ICD-9 codes for heart disease (410-414 and 420-429).35 Other important potential unmeasured confounders that are not available in electronic medical records include behavioral or environmental risk factors, such as poor diet, low level of physical activity, or exposure to second-hand smoke, although we have no reason to believe that patients with these risk factors would be more likely to be prescribed metoprolol tartrate rather than atenolol.
In conclusion, we found no differences in CV event rates when comparing patients without a history of CV events who were initiating treatment with either atenolol or metoprolol tartrate. These findings suggest that hypertension trial outcomes with atenolol may not relate to unfavorable characteristics of this particular drug. These results should be interpreted cautiously, since there have been no trials comparing these 2 β-blockers directly.
Correspondence: Emily D. Parker, MPH, PhD, HealthPartners Institute for Education and Research, Box 1524, Mail Stop 21111R, Minneapolis, MN 55440-1524 (Emily.D.Parker@Healthpartners.com).
Accepted for Publication: June 11, 2012.
Published Online: August 27, 2012. doi:10.1001/archinternmed.2012.4276
Author Contributions:Study concept and design: Parker, Margolis, and O’Connor. Acquisition of data: Margolis, Trower, Magid, Tavel, Shetterly, Swain, and O’Connor. Analysis and interpretation of data: Parker, Margolis, Ho, and O’Connor. Drafting of the manuscript: Parker, Margolis, Trower, Swain, and O’Connor. Critical revision of the manuscript for important intellectual content: Parker, Magid, Tavel, Shetterly, and Ho. Statistical analysis: Parker and Trower. Obtained funding: Magid and O’Connor. Study supervision: Margolis.
Financial Disclosure: None reported.
Funding/Support: This project was funded by grant NIH/NHLBI/U19 HL091179 from the National Heart, Lung, and Blood Institute and subcontract to HealthPartners Institute for Education and Research.