Selective β₁-Antagonism with Bisoprolol Is Associated with Fewer Postoperative Strokes than Atenolol or Metoprolol: A Single-center Cohort Study of 44,092 Consecutive Patients

THE initial enthusiastic use of perioperative β-blockers, which was largely based on two positive trials,^1,2  has been tempered after the publication of the results of the Perioperative Ischemia Evaluation (POISE) trial.³  POISE trial demonstrated that although metoprolol succinate reduced the risk of perioperative myocardial infarction, it also doubled the incidence of postoperative stroke and increased mortality. This finding was also confirmed by subsequent meta-analyses that showed a similar reduction in myocardial infarction and increase in stroke rates in all perioperative trials that used either formulation of metoprolol (tartate and succinate) or atenolol.⁴ 

The specific reasons why perioperative β-blockade may cause acute stroke have not been elucidated. A potential mechanism is impairment in normal physiological responses to significant perioperative blood loss, which is also associated with postoperative stroke.³  Notably, the sequelae of major surgical blood loss, namely acute perioperative anemia, is both common⁵  and entails adaptive homeostatic responses that include increases in cardiac output,⁶  cerebral blood flow, and cerebrovascular dilatation.^7,8  Less selective β-antagonists have the potential to impair all of these adaptive responses.

To further explore this potential mechanism for perioperative stroke, we undertook a series of animal experiments to investigate the separate roles of β₁- and β₂-adrenergic receptors in mediating the physiologic response to acute anemia. In a rat model of acute anemia, the use of a weakly β₁-selective β-blocker metoprolol (relative β₁/β₂ affinity of 2.3) impaired the cardiac output response and reduced brain tissue Po₂, thereby causing global cerebral hypoxia.⁹  In addition, the specific antagonism of the β₂-receptor in the setting of acute anemia attenuated β₂-mediated cerebral vascular dilation¹⁰  and reduced brain Po₂.¹¹  In comparison, treatment with a low dose of a highly β₁-selective β-blocker, nebivolol, effectively inhibited the cardiac response to anemia without attenuating brain Po₂, possibly by preserving the vascular response to acute anemia.¹² 

These experimental data suggest that the relative β₁-selectivity of β-blockers may influence their efficacy in preventing postoperative cardiovascular outcomes, as has been suggested by a recent meta-analysis.¹³  Notably, the three most commonly used chronic β-blockers in clinical practice vary with respect to their β-receptor selectivity. Specifically, their relative β₁/β₂ affinities are: 2.3 for metoprolol, 4.7 for atenolol, and 13.5 for bisoprolol.^14,15 

We therefore undertook this retrospective cohort study to evaluate the association of β-blocker selectivity with the risk of postoperative acute stroke. Our primary hypothesis was that a more β₁-selective agent, namely bisoprolol, would be associated with lower adjusted risk of postoperative stroke than the less selective antagonists such as metoprolol or atenolol.

We conducted a single-center, retrospective cohort study on patients having intermediate- to high-risk noncardiac surgery at the University Health Network in Toronto, Ontario, Canada. The study conformed to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines. The University Health Network, which is a teaching hospital network affiliated to the University of Toronto (Toronto, Ontario, Canada), offers a full range of adult noncardiac surgery performed at three sites. This study was initiated after the approval by the University Health Network Institutional Research Ethics Board (Toronto, Ontario, Canada), which waived the requirement for informed consent for this study.

We evaluated consecutive patients who were aged 50 yr or older, had American Society of Anesthesiologists physical status scores of four or less, underwent noncardiac and nonneurologic surgery from January 2003 to December 2010, and required an overnight admission to hospital. We also excluded patients who underwent solid organ transplantation. To define the timing of each stroke, the first surgery was considered as the index in patients having more than one surgery during the admission.

All perioperative data were prospectively collected in linked institutional databases, as has been previously described.^5,16,17  These data included demographics, comorbidities, medications, surgical information, laboratory test results, blood product transfusion, and postoperative events. The accuracy of the database has also been confirmed. In brief, we compared all the relevant electronic data with the paper record in 100 randomly selected patients and found an error rate of less than 2%.¹⁶  The type and dose of β-antagonists were recorded and for the most part reflects chronic longer term use.

Data on β-blocker exposure was defined primarily using two separate databases, the electronic preoperative assessment data, which is a surrogate for chronic use but does not include data for emergency surgery or inpatients transferred from other hospitals before elective surgery. Approximately 35% of all surgical patients have no data on preoperative β-blocker exposure. The inpatient pharmacy database shows all inpatient drug use but cannot distinguish between acute and chronic use or those withdrawn from β-blockers. Previous work from our groups shows that approximately 30% of patients are started on β-blockers between the inpatient visit and surgery, whereas less than 10% are withdrawn from β-blockers, when the preoperative data are compared with the inpatient pharmacy database. Thus the current analysis is based on in-hospital exposure to β-blockers that is predominately chronic in nature. The β-blocker exposure in this analysis reflects data from the inpatient pharmacy database for each of the first six postoperative days, where the day of surgery is considered the day 1. To establish the consistency of the β-blocker dosage and type and dose, we compared the drug administered on the first 2 postoperative days to the rest of the 6 postoperative days and establish the β-blocker crossover rate, in each β-blocker exposed patient. Less than 1% (107 of 10,756) of all patients received different β-blockers between the first and last inpatient days. We included all of these patients in the analysis because this small number of patients was unlikely to influence the outcome. The drug exposure of interest was considered to be the first β-blocker administered in all cases.

The primary outcome was postoperative stroke within 7 days of surgery, and this definition was defined a priori because POISE trial showed the increase in the number of strokes in patients receiving metoprolol, compared with the control group occurred in this time frame.³  Identification of patients with stroke was defined in a multiple step process. First, we surveyed the discharge abstract database of 10th version of the International Classification of Diseases-10 (ICD-10) codes for cerebral ischemia and stroke (appendix). We then cross-referenced these results with the medical imaging database for patients having computed tomography or magnetic resonance imaging of the brain within 7 days of surgery. After reviewing the medical and radiological reports of this cohort, two investigators (Drs. Judelman and Ashes) adjudicated the presence of postoperative strokes based on the presence of confirmatory clinical or radiological evidence. The stroke timing and adjudication occurred without knowledge of medication history or hematologic status. Finally, we surveyed all patients who died within 7 days of surgery to reduce the number of missed strokes in patients who died before diagnosis. All disagreements were resolved through consensus or involvement of a third reviewer (Dr. Beattie).

The secondary outcome was a composite comprising in-hospital death within 30 days of surgery, stroke, and postoperative myocardial injury, which was defined as a peak troponin I of greater than 0.7 μg/l. Myocardial injury was assessed using troponin I, using the Dade Behring Dimension assay (detection limit 0.07 μg/ml; Siemens Healthcare Diagnostics, Deerfield, IL). In May 2008, the institution changed to the Abbott Architect i2000 analyzer (Abbott Diagnostics, Abbott Park, IL). Internal correlation studies showed a 1-to-1 correlation to the Dade Behring assay and the change in assay method does not affect the interpretation of the results presented here.

We compared the demographic, surgical, comorbid, and pharmacologic characteristics of the patients taking the different β-blockers with patients who were not exposed to in-hospital β-blockers. We then assessed the unadjusted relationship between nadir hemoglobin^†† and postoperative stroke by constructing a restricted cubic spline curve for each of metoprolol, atenolol, and bisoprolol. Because the number of strokes limited the number of confounding variables that could be reliably entered into logistic regression analysis, the primary analysis consisted of a propensity score-matched pairs analysis comparing bisoprolol with the combination of metoprolol or atenolol. A propensity score estimating the probability of being exposed to bisoprolol was generated using a nonparsimonious multivariable logistic regression model that included all measured covariates.^‡‡ Individuals receiving bisoprolol were then matched to individuals receiving metoprolol or atenolol, on a 1-to-1 basis. In this analysis, two cohorts with similar characteristics were created using a 5 to >1 computerized greedy-matching algorithm. Covariate balance between the matched pairs was then evaluated using standardized difference. This process was repeated iteratively, blinded to the outcomes, until good covariate balance (standardized difference <10%, in every confounding variable) was achieved. We used conditional logistic regression to compare rates of the primary outcome and secondary outcome, in the matched pairs.

A second analysis was then conducted using multivariable logistic regression analysis. The candidate variables in this model included the predictors of stroke identified in the POISE trial and two potential predictors of stroke that were identified based on our animal experiments: a nadir hemoglobin under 9 gm/dl and β-blocker type. In this analysis, β-blockers type compared bisoprolol with the two agents, metoprolol or atenolol. Model calibration was evaluated by the Hosmer–Lemeshow test, whereas discrimination was assessed using the c-statistic. All analyses were performed using SAS 9.2 (SAS Institute, Cary, NC)

The final cohort consisted of 44,092 consecutive patients having in-hospital procedures. The details of the assembly of this cohort are presented in figure 1, whereas the characteristics of the study sample are presented in table 1. Eighty-eight patients (0.2%) experienced acute stroke within 7 days of surgery. Among patients with a Revised Cardiac Risk Index of 1 or more, the risk of stroke was 0.4%. The peak incidence of stroke was on the first postoperative day, and approximately two thirds of strokes occurring within 48 h of surgery (fig. 2). Of the 88 patients who suffered postoperative strokes, 17 (19.3%) died in hospital within 30 days of surgery. The median length-of-stay of the surviving stroke patients was 27 days (intraquartile range 14–47), contrast to 4 days (intraquartile range 2–8) for those patients who survived without having a stroke. Of the surviving stroke patients, 57 (80%) required long-term care, rehabilitation facilities, or some form of extra care on hospital discharge. In comparison, this proportion was 33% among surviving nonstroke patients.

Twenty-four percent (n = 10,756) of the patients in this study sample were taking β-blockers during the hospital stay. The patient characteristics, including demographics and comorbidities, the median dosage and range of each of the β-blockers of interest, surgical details and transfusion requirements, and concomitant cardiovascular medications are seen in table 1. It is important to note that patients taking atenolol had fewer risk factors associated with stroke than patients receiving bisoprolol or metoprolol. Specifically, patients receiving atenolol had less cerebral vascular disease, chronic renal failure, and diabetes. The patients exposed to atenolol were being predominately treated for hypertension (57%).

The relationship between nadir hemoglobin and stroke is displayed in figure 3. There is a difference in the baseline risk of stroke between metoprolol and the other two agents. Second, there is a sharp increase in the incidence of stroke when hemoglobin levels decrease to below a level of 9 gm/dl.

The matched cohort consisted of 2,462 patients; half receiving bisoprolol and the remainder receiving either metoprolol or atenolol. The pairs were well matched for all variables. The mean standardized difference for the 50 known variables was −0.008 and ranged from −0.098 to 0.045 (table 2). In the matched cohort analysis, two patients taking bisoprolol and 10 patients taking less selective β-antagonists experienced postoperative stroke (odds ratio, 0.20; 95% CI, 0.04–0.91). The components of the composite secondary outcomes showed less acute myocardial injury as judged by increased troponin after surgery, and no difference in mortality.

The result of the restricted logistic regression analysis comparing bisoprolol with atenolol and metoprolol is seen in table 3. Bisoprolol is associated with fewer postoperative strokes than the less selective β-blockers (0.20, 95% CI, 0.04–0.99). In addition, factors associated with cerebral hypoxia and stroke from earlier animal experiments and clinical studies were retained in the current study. Acute anemia was associated with stroke in a manner that was dependent on the magnitude of the decrease in hemoglobin concentration. The most significant predictor of stroke was a history of cerebral vascular disease. New onset of atrial fibrillation (AF), an important factor in POISE trial, was not significant in our analysis (odds ratio, 0.72; 95% CI, 0.1–5.3). This model has good discrimination (c-statistic, 0.788) and good calibration (P = 0.712).

In this single-institution retrospective cohort study, postoperative stroke was a rare but devastating complication of major noncardiac surgery. The highly selective β₁-antagonist bisoprolol was associated with decreased stroke rate compared with the less β₁-selective agents metoprolol or atenolol. The increase in stroke rate observed was also associated with an increase in other severe adverse clinical outcomes suggesting that the impact of the less selective agents have broad clinical effects on patient outcomes. This analysis also demonstrates that there is an increase in adverse cardiac outcomes when β-blocked patients experience acute anemia (hemoglobin <9 gm/dl).

Our results warrant comparison with the POISE trial³  and our animal experiments^9–12 . Similar to the POISE trial, cerebral vascular disease was a major risk factor for postoperative stroke. A history of cerebral vascular disease or stroke is the most consistent predictor of postoperative stroke and has been seen in several recent large observational trials.^17–19  In addition, transfusion was associated with stroke in a dose-dependent manner, which has been previously demonstrated.²⁰  In that analysis, 4 units of transfusion were used to define major blood loss as compared with 2 units in POISE trial. Our analysis, which is composed of predominantly of patients with chronically administered β-antagonists, also demonstrates that metoprolol is associated with an increased incidence of postoperative stroke, as has recently been observed in two other retrospective cohort studies.^21,22  In those investigations, metoprolol was compared with atenolol, however, we find these two drugs were essentially equivalent in increasing the stroke rate compared with unexposed patients (analysis not shown). The differences between metoprolol and atenolol in those analyses^5,22  could result from confounding by indication as was discussed.²¹  Our data supports this supposition given that, in our study, atenolol-exposed patients had much lower risk of stroke. The baseline risk factors for stroke in patients administered metoprolol and bisoprolol were similar limiting the risk of confounding by indication. These findings that bisoprolol may be a safer alternative in β-blocked patients clearly warrants further investigation in a controlled clinical trial.

Our experimental findings suggested that the risk of stroke associated with less selective agents results from inhibition of β₂-mediated cerebral vasodilation. Animal studies show that all the β-blockers evaluated would be expected to have a comparable effect on limiting the cardiac output responsiveness through comparable antagonism of β₁-receptors.^9–12  The compensatory increase in cardiac output becomes predominantly heart rate dependent at a hemoglobin level between 9 and 10 gm/dl.⁶  Below this hemoglobin level, all of the β-blockers that we studied were associated with increasing stroke rates (fig. 3). However, as previously described, the impact of β₂-inhibition by less selective β-blockers may impair cerebral vasodilation independent of hemoglobin level; possibly explaining the difference seen between metoprolol and bisoprolol at nadir hemoglobin levels greater than 9 gm/dl (fig. 3) because nonselective β-blockers may impair cerebral vasodilation (β₂-mediated effect) across all levels of blood loss. Our animal data support this argument by demonstrating that metoprolol increased systemic vascular resistance and reduced brain tissue Po₂in vivo, at normal hemoglobin levels.¹⁰  In addition, metoprolol clearly impaired β₂-adrenergic vasodilation of cerebral arteries in vitro.¹⁰  Finally, systemic β-blockers administered to anemic animals, at doses that can interact with β₂ receptors, showed acute reduction in brain tissue Po₂ and increased expression of biological markers of cerebral hypoxia.^8–10 

Kamel et al. first noted an association between acute anemia and postoperative strokes. Using the National Surgical Quality Improvement Project database, they found that transfusion of more than 4 units doubled the stroke rate. Two additional, and independent, studies document that the interaction between β-blockade and acute surgical anemia is associated with an increased risk of postoperative coronary morbidity and multiorgan failure.^5,23  Because selective β₁-antagonists preserve peripheral circulation, it follows that highly cardioselective agents favor vital organ perfusion during conditions of acute blood loss and fluid resuscitation. This is a possible explanation for the finding that chronically administered metoprolol was associated with a greater risk of death after surgery compared with more selective β-antagonists.²⁴ 

The overall stroke rate in the current study was 0.2%. In comparison, in studies where the National Surgical Quality Improvement Project database was used to study acute postoperative stroke, a range of stroke incidence from 0.1 to 0.4% was observed. Although our incidence of stroke falls within this range, it is difficult to explain a four-fold variation in stroke rate, in studies using the National Surgical Quality Improvement Project database but may be due to different study populations. In the subgroup analysis of our sample, patients with at least one cardiac risk factor had an incidence of 0.4%, doubling the overall stroke incidence.

Previous studies have identified numerous risk factors for perioperative stroke which have included age, sex, AF, congestive heart failure, renal disease and cardiovascular disease,¹⁷  chronic obstructive pulmonary disease, left ventricular systolic dysfunction, duration of surgery, and hyperglycemia. Our study did not have the sample size or the power to assess these factors. However, they are unlikely to influence our results because they were well matched in our propensity score. Studies that identified these variables had on average more than 1,000 outcomes and more than 250,000 subjects.

β-Antagonism is known to decrease postoperative AF as demonstrated by the POISE study. Paradoxically, metoprolol doubled the stroke rate in POISE trial. These divergent data suggest that the postoperative strokes associated with β-blockade are unlikely to be related to new onset AF. Furthermore, the incidence of new onset postoperative AF in our study was comparable with POISE trial. However, we did not find that AF was associated with stroke. The potential explanations for this include: (1) AF may have been underestimated or AF may not have been coded properly in the discharge abstract database if it was of short duration or did not require special therapy; (2) previous associations between AF and postoperative stroke were derived on the basis of a history of preoperative AF,^18,19 which is distinct from our study and POISE trial, which found that new onset of AF was associated with postoperative stroke; (3) the American Heart Association guidelines for the management of AF suggest the use of β-blocker alone or in combination with other drugs.²⁵  This guideline may have contributed to the previous associations between AF and stroke because over 60% of patients with new onset AF reportedly receive metoprolol.²⁶  Unfortunately, studies finding that preoperative AF was associated with postoperative stroke did not record details of β-blockers use and they did not adjust for the influence of other chronic medication. Thus, the degree to which previous β-adrenergic antagonism confounds these results remains unknown because no difference in AF prophylactic effects were identified when different β-antagonists were prescribed.²⁷ 

The retrospective observational nature of this study can be influenced by potential unmeasured confounders and the two factors, β-adrenergic antagonism and acute anemia may identify patients who are “sicker,” having procedures with increased complexity. Nonetheless, using a propensity score-matched cohort, we were able to balance all known confounders and show a reduction in the stroke rate with highly selective β₁-antagonism. Cautious interpretation is also warranted because there were less than 90 strokes, rendering the results statistically fragile. However, the odds ratios found by our targeted logistic regression and propensity-matched cohorts are similar. In our opinion, our results are strengthened by the fact that two methods of risk adjustment, using different assumptions, yield similar results.

The use of International Classification of Diseases-10 coding to identify postoperative stroke may have introduced some errors, as other neurological pathology is included in the acute stroke coding.^28,29  The strokes we have identified in this analysis were independently verified by two observers, blinded to β-blocker exposure, using both clinical and radiological evidence. We are unable to comment on the clinically important incidence of perioperative hypotension^3,30  because intraoperative hemodynamic data on all patients were not available. Acute anemia is strongly associated with hypotension. The Transfusion Trigger Trial for Functional Outcomes in Cardiovascular Patients Undergoing Surgical Hip Fracture Repair trial showed that a transfusion trigger of 8 gm/dl was associated with a three-fold increase in serious hypotension, tachycardia, and congestive failure compared with the cohort with a higher transfusion trigger.³¹ 

We cannot be sure that we have accurately defined all β-blocker exposures; in urgent and emergent procedures data are based solely on inpatient pharmacy data. This limits us because there is the potential to label a patient who had β-blockers withdrawn, as not being exposed or alternatively drug newly started, as a result of some event, could be labeled as chronic use. Both of these scenarios carry different risks.³²  Finally, the use of bisoprolol increased substantially over the period of our study. Although we have balanced the number of patients by year in the propensity analysis, we cannot adjust for other factors that may have influenced the decision to use bisoprolol over this time period.

Postoperative stroke is a rare but devastating event that is associated with the use of the nonselective β-blockers and acute anemia. The analysis further identifies two potentially modifiable risk factors, the type of β-blocker and anemia. Consideration of these factors in the management of at-risk patients may alter the safety profile of perioperative β-adrenergic antagonism which offers irrefutable perioperative cardioprotection. All studies to date show that patients with significant cerebral vascular disease are at risk of stroke and treatment with β-antagonists, of any kind, should be carefully considered. Patients needing β-antagonist-mediated cardioprotection during surgery may benefit from the use of highly selective β₁-agents. We are in the process evaluating these findings prospectively.