We identified consecutive patients undergoing elective surgery in any hospital throughout Ontario, Canada, by using the database of the Canadian Institutes for Health Information, which provides the official data for ongoing accreditation and financial reimbursement in this setting.49
The accrual period was from 1 April 1992 to 1 April 2002 (10 years), representing all years available for analysis. We identified elderly patients (older than 65) admitted to hospital for elective surgery. We did not include outpatients, patients having surgery as a day procedure, or young individuals because of the generally low event rates in such circumstances. To reduce confounding from differing amounts of pre-existing illness,50
we excluded in advance patients with symptomatic coronary disease as evidenced by chronic use of nitrates.51
We made special efforts to gather population based data that eliminated referral bias and counted each patient only once. We deleted duplicate hospital records by retaining only one copy of events characterised as the same admission for the same patient on the same day. We analysed only the first admission for patients with more than one elective surgical procedure during the study interval (results based on separate admissions yielded more extreme results and are not reported). In addition, we counted outcomes after transfers according to the hospital first involved. We used confidentiality safeguards at the Institute for Clinical Evaluative Sciences in Ontario to conduct the study. All databases have been used extensively in past research.52-54
For each patient we searched previously validated, population wide prescription records for the year before admission,55,56
reasoning that β blocker medications would probably be continued in the perioperative setting. We classified individual patients who received two or more prescriptions for atenolol as using this medication on an ongoing basis. Similarly, two or more prescriptions for metoprolol identified patients who used that β blocker on an ongoing basis. We classified patients receiving two or more prescriptions for both atenolol and metoprolol as having mixed exposures and reported separately. We lacked direct individual data on medications received in hospital; however, to validate our classifications further we also identified prescriptions after discharge to confirm ongoing use of either atenolol or metoprolol among survivors.
We also considered some more complicated situations. We examined patients receiving two or more prescriptions of carvedilol, labetolol, oxprenolol, pindolol, timolol, or acebutolol in a secondary analysis of other short acting β blockers. Similarly, we examined patients receiving two or more prescriptions for nadolol or bisoprolol in a secondary analysis of other long acting β blockers. We considered patients receiving sustained release formulations of any β blockers in the category of other long acting β blockers and patients receiving multiple prescriptions of different short acting and long acting β blockers a mixed group. We considered patients receiving two or more prescriptions for sotalol or propranolol each a unique group because of the distinct indications for these particular β blockers.
Outcome and characteristics
We obtained information on patients' demographics by linking individuals to the Registered Persons database, the official governmental record for patients in Ontario. We obtained information on the nature of the surgical procedure and postoperative recovery from the Canadian Institutes for Health Information database. In addition, we classified operations according to type of surgery as either cardiac or non-cardiac, with non-cardiac surgery further distinguished as high risk noncardiac (thoracic, abdominal, retroperitoneal, vascular), medium risk non-cardiac (neurosurgical, external head and neck, unclassified), and low risk non-cardiac (lower urological and gynaecological, orthopaedic, breast and skin, ophthalmological). The available databases did not contain data on compliance, family history, or lifestyle.
We defined the primary outcome as death or myocardial infarction occurring during hospitalisation.57
Secondary analysis examined each end point separately. We also analysed non-cardiac complications after surgery to check for a lack of differences where no differences would be anticipated. These additional analyses of seven tracer conditions included wound infection, ileus, pneumonia, aspiration, respiratory failure, renal failure, and delirium. We further tested comparability between the two groups by examining five pre-specified, distinct, available measures of the process of care: transfusion of blood component; ultrasound of the abdomen; accidental cuts, puncture or perforation; foreign object left in body; and failure of dosing, instrument, or sterilisation.
We used the χ2
test to compare the frequency of death or myocardial infarction for patients receiving atenolol and metoprolol, because the data lacked the day of the postoperative infarction. We also used the log rank statistic to analyse dates of death alone. Additionally, we constructed a general predictive model by subjecting each baseline patient factor to stepwise logistic regression and thereby obtained an adjusted comparison of rates of death or myocardial infarction for patients receiving atenolol or metoprolol. A multivariable propensity score analysis,58,59
designed to adjust for clinical determinants of β blocker selection, yielded results almost identical to those based on the general predictive model and are not reported (results same to two decimal places). All P values were two tailed, estimates calculated with 95% confidence intervals, and analyses conducted by using SAS software (version 8.02, Cary, NC 27513, USA).
A total of 454 336 elderly patients had a total of 634 925 admissions for elective surgery across 252 separate hospitals during the 10 year interval. A minority of patients had symptomatic coronary disease (n = 48 128), and most (n = 345 253) had not received a β blocker on an ongoing basis as an outpatient in the year before surgery. The most commonly used β blockers were atenolol (n = 23 091) and metoprolol (n = 14 060), forming a ratio of about 5:3 that was stable over the decade. The median dose of atenolol was 50 mg once daily and that of metoprolol 50 mg twice daily. Some patients received another long acting β blocker (n = 2754), some another short acting β blocker (n = 10 668), and some had mixed exposures to a long acting as well as a short acting β blocker (n = 229). The remaining patients were those receiving sotalol (n = 3810), propranolol (n = 6309), or either of these two medications in combination with another β blocker (n = 34).
As expected, the baseline characteristics for patients receiving atenolol or metoprolol overlapped substantially (). The largest difference was in the proportion who had cardiac surgery, which was more common among those receiving metoprolol, although it was generally infrequent in both groups (an imbalance examined in subsequent analyses). After excluding patients undergoing cardiac surgery, the atenolol and metoprolol groups were remarkably similar in use of cardiac medications including statins, digoxin, furosemide, calcium channel blockers, angiotensin pathway blockers, and anticoagulants (see bmj.com
for characteristics of patients not having cardiac surgery). We found no major differences between the two groups in other medications used to treat chronic medical and psychiatric conditions. We also found no clinically important differences in demographic characteristics between the two groups.
Patients' characteristics. Values are numbers (percentages) of patients
A total of 1038 patients experienced a myocardial infarction or died during their stay in hospital. The risk of this combined end point was one fifth lower for patients receiving atenolol rather than metoprolol (2.5% v 3.2%, P < 0.001). The difference in risk was also apparent for the solitary end points of myocardial infarction (1.6% v 2.0%, P = 0.004) and of death (1.2% v 1.6%, P = 0.007). The difference persisted in those patients not having cardiac surgery, both for the combined end point (2.0% v 2.6%, P < 0.001), and the solitary end points of myocardial infarction (1.1% v 1.4%, P = 0.024) and death (1.2% v 1.6%, P = 0.003). The pattern was consistent for high risk, medium risk, and low risk non-cardiac surgery and not apparent with cardiac surgery (). The relative risk reduction persisted in subgroups that excluded those patients receiving calcium channel blockers, furosemide, or other single cardiac medications.
Fig 1 Relative reduction in risk of death or myocardial infarction. Relative risk of myocardial infarction or death for patients receiving atenolol compared with patients receiving metoproplol. A x axis value of 0 denotes the null effect, where risk with atenolol (more ...)
We constructed a clinical prediction rule by taking into account each patient's baseline characteristics, concurrent medications, and type of surgery. The important independent predictors of myocardial infarction or death were the patient's age and sex; four medications (furosemide, calcium channel blockers, angiotensin converting enzyme (ACE) inhibitors, and statins); and type of surgery (). The overall goodness of fit of this model was moderate (area under the receiver operating characteristic curve 0.74, P < 0.001) and similar to past published perioperative prediction rules (area under the curve 0.60-0.65).60
The difference between atenolol and metoprolol persisted after adjusting for these predictors (relative risk reduction 13%, 95% confidence interval 1% to 22%).
Independent predictors of myocardial infarction or death
We conducted two further tests to gauge the robustness of our findings. A comparison of any long acting β blocker (not just atenolol) to any short acting β blocker (not just metoprolol) showed a 15% reduction in risk of myocardial infarction or death (5% to 24%) before adjusting for the predictors and a 10% risk reduction after adjusting for the predictors (0% to 19%). A comparison of those with confirmed ongoing use of atenolol or metoprolol (both defined as two or more prescriptions for the corresponding medication in the year after surgery) yielded a 45% reduction in risk of myocardial infarction (31% to 56%) before adjusting for the predictors and a 35% risk reduction after adjusting for the predictors (19% to 49%).
We observed no differences between atenolol and metoprolol when we examined non-cardiac outcomes and processes of care after surgery that might be related to unmeasured characteristics of patients, surgical procedures, or hospitals. Postoperative wound infection was the most common complication and showed no significant difference between the two groups (). Delirium was the second most common complication with a slight imbalance against the metoprolol group that was small in magnitude and did not reach significance (P = 0.15). Postoperative pneumonia, renal failure, and prolonged ileus were all similar in frequency (each about 1%), and the two groups did not differ significantly. Misadventures were rare, with no consistent imbalance between the two groups.
Non-cardiac postoperative outcomes. Values are numbers (percentages) of patients unless otherwise indicated
Most deaths occurred soon after admission, with postoperative day 3 as the most common. Differences between atenolol and metoprolol were primarily observed from postoperative days 2 through 14, in keeping with acute cardiac stress after surgery (). Differences between atenolol and metoprolol were not evident on the day of admission, in keeping with intraoperative catastrophes. Differences between atenolol and metoprolol were not evident beyond day 14, in keeping with delayed non-cardiac complications (and not easily explained by a selection bias that would entail an inherent ongoing difference in risk). No day showed a significant difference in mortality that favoured metoprolol. Analyses based on comparing any long acting β blocker to any short acting β blocker showed similar patterns.
Fig 2 Absolute risk of death in hospital within 30 days of elective surgery. Absolute risk of death comparing atenolol with metoprolol after surgery. Data expressed as cumulative number of deaths per 1000 admissions on corresponding day. P values compare death (more ...)