β-Blockers after Myocardial Infarction: Influence of First-Year Clinical Course on Long-Term Effectiveness

Methods

Patients

Between 19 June 1978 and 2 October 1980, the Beta-Blocker in Heart Attack Trial (BHAT) enrolled 3837 patients in 31 clinical centers in the United States and Canada. Eligible patients included men and women between the ages of 30 and 69 years who had been hospitalized with an acute myocardial infarction that met traditional clinical and laboratory criteria. Patients were randomly assigned to receive propranolol or matching placebo after their condition had stabilized in the hospital, between 5 and 21 days after admission. (A dose of 180 or 240 mg/d was assigned at 4 weeks, depending on serum response to an initial dose of 120 mg; placebo recipients were also assigned 180- or 240-mg dose schedules to maintain the study's double-blind design.)

Follow-up clinic visits, which included physical examinations and patient interviews, were scheduled at 4 and 6 weeks, and every 3 months after randomization until the termination of the trial. The minimum length of follow-up was 12 months, and the average time in the trial was 25 months. (Complete descriptions of the BHAT design and methods have been previously reported [10, 11].)

To classify patients by treatment status after 12 months, as well as to ensure adequate data on clinical course during the first year, the cohort for the present study was restricted to patients with a completed physical examination at the 12-month visit and at least four of six possible follow-up clinical examinations done during the first year.

Of the 3647 patients known to have survived at least 360 days from the date of randomization, we excluded 308 (8%) who did not have a physical examination at the scheduled 12-month visit and 42 (1%) who had three or fewer visits. The resulting cohort had the same sex, race, and age distribution as the original BHAT cohort. However, 20% of the study cohort had been withdrawn from their randomly assigned treatment by the 12-month visit. Of the 1666 patients started on propranolol at the original zero-time, 20% had been withdrawn: One hundred forty-three (9%) had been started on open-label β-blockers, and 187 (11%) had been withdrawn from all β-blocking medications. A similar proportion of the 1631 patients originally started on placebo (21%) had been withdrawn as of 1 year after randomization: One hundred thirty-seven (8%) had been kept off all β-blockers, and 204 (13%) had been started on active treatment. At the 12-month visit, therefore, 1683 patients were receiving β-blocker therapy, 98% of whom were taking propranolol (mean dose, 167 mg/d), and 1614 patients were not receiving β-blocker therapy (79% were taking a placebo).

Specification of Clinical Course during Year 1 of the Trial

Information on each patient's status regarding four clinical dimensions during the first 12 months of the trial were derived from the structured interviews and physical examinations administered at each follow-up clinic visit during the first year. The four dimensions (worsening myocardial ischemia, congestive heart failure, major cardiac arrhythmias, and noncardiac comorbid conditions) were selected on the basis of previous research that has established them as major pathways by which mortality is increased in the postinfarction period [12]. The specific clinical features defining these dimensions are described below.

Worsening Myocardial Ischemia

The indicators for worsening myocardial ischemia were as follows: angina, worsening angina, angina at rest or daily, and recurrent infarction. Each indicator was classified by its degree of severity as follows: absent (symptom or event was not reported in any study interval in the first year); mild (feature was present in less than 50% of the study intervals); and severe (feature was present in more than 50% of the study intervals).

Congestive Heart Failure

The indicators of congestive failure were as follows: symptoms of shortness of breath or dyspnea; identification of pulmonary rales or S₃ gallop on physical examination; and overt episodes of congestive failure. These indicators were further classified by the proportion of study intervals in which each feature was reported as described previously for worsening ischemia.

Major Cardiac Arrhythmias

Evidence of electrical instability meant that at least one of the following conditions was recorded during the first 12 months: atrial fibrillation or flutter, supraventricular tachycardia, third-degree atrioventricular block, ventricular tachycardia, and ventricular fibrillation.

Major Noncardiac Comorbid Conditions

We used a modification of the Charlson comorbidity index [13], assigning points for the presence (at baseline or any time during the first 12 months) of several conditions: 1 point each for peripheral vascular disease, cerebrovascular disease, pulmonary disease, and diabetes and 2 points for renal disease. A composite score was then calculated for each patient.

Follow-up Risk Strata: A More Refined Specification of Clinical Course

Clinical courses in the first year were identified separately for each of the four clinical dimensions according to strata of increasing risk. For a given dimension and feature within it, mortality rates (to 2 April 1982, the last date of follow-up) were calculated for patients with and without the feature during the first year, regardless of the patients' β-blocker treatment status. The feature correlating with the highest mortality rate was taken to define the worst clinical course or stratum in that dimension. The next worst course was then identified by calculating mortality rates for the remaining features after removing patients in the worst risk stratum. This process was continued until an exclusive, hierarchical progression of risk was defined for each clinical dimension. The resulting clinical courses of increasing risk are shown in Table 1.

The final risk strata were constructed by sequentially combining and consolidating these individual clinical courses. Patients were first cross-classified by their clinical courses for worsening ischemia and congestive heart failure. Clusters of cells in the cross-classified tables were consolidated into new risk strata on the basis of the similarity of mortality rates among the cells. The additional contributions of cardiac arrhythmia and noncardiac comorbidity to prognosis were incorporated into the risk system by an analogous process of further cross-classification and consolidation. Technical details of the consolidation method have been reported previously [14].

The final risk strata, which may also be viewed as a more refined specification of each patient's clinical course, are defined in Table 2. Each course is interpreted by referring to the component clinical features noted for each line in the table. For example, patients were classified in stratum I if during the first year they had not experienced a recurrent myocardial infarction (although anginal symptoms may have been present), had no evidence of congestive heart failure (signs, symptoms, or overt episodes), and did not have severe comorbid disease (at most, one condition could be noted). The presence or absence of major cardiac arrhythmias did not alter the patient's risk classification at this level. Patients were classified in stratum II if they had no evidence of electrical instability or severe comorbidity but did have the indicated combinations of worsening ischemia and congestive heart failure (for example, if a recurrent myocardial infarction had occurred, only symptoms, but no signs or overt episodes, of congestive heart failure could be present; if a probable or confirmed episode of congestive failure had occurred, no evidence of worsening ischemia could be present). Patients were classified at highest risk (stratum III) when more serious combinations of congestive failure, ischemia, and cardiac arrhythmias were present or when the patients had two or more comorbid conditions.

Treatment Status and Baseline Comparison by Clinical Course

Patients were classified as being on treatment at 12 months after randomization if they were receiving β-blocker therapy at the time of the 12-month visit. Patients were considered off treatment if they were not receiving β-blocker therapy at that time. Baseline demographic and clinical features (defined at the original zero-time for the trial) of the study population are presented by treatment status at 12 months for each of the three final risk strata in Table 3.

Statistical Analysis

The life-table program (1L) of the BMDP statistical package was used for the calculation of Kaplan-Meier survival curves [15] within each risk stratum, overall and by treatment group. Chi-square statistics and 95% CIs were calculated as indicators of the strength of estimated effects. Relative risks were computed from the ratio of Kaplan-Meier survival rates after 12 months (evaluated at 720 days of follow-up); lower and upper bounds for the relative risk were estimated using the standard errors from the survival curves [16]. Overall differences in survival curves were assessed by calculation of the log-rank test statistic [17].

Results

Unstratified Mortality Rates

In the first year of the trial, 70 deaths (3.7%) occurred among the 1916 patients originally started on propranolol, and 115 deaths (6.0%) occurred among the 1921 patients originally assigned to placebo. The risk reduction of 39% was both clinically impressive and statistically significant (95% CI, −54% to −18%). In contrast, the estimated mortality rate after the first year was 5.6% in patients on β-blocker treatment compared with 7.2% for those off treatment at 1 year. The proportional risk reduction of 22% was less impressive clinically and was not indicative of a statistically significant difference (95% CI, −43% to 8%).

Stratified Mortality Rates

The crude risk reduction of 22% after year 1 does not take into account differences in underlying risk between treatment groups at 1 year. Using the three risk strata defined at 12 months, we calculated severity-adjusted mortality rates for patients who were on (5.6%) and off (6.9%) β-blocker treatment. Table 4 summarizes the effects of β-blocker treatment at different follow-up points in the BHAT trial. For the overall trial analyzed by randomized treatment group with an average follow-up of 25 months, β-blocker therapy resulted in a 27% reduction in the risk for death. For the first year of follow-up, the risk reduction was even larger (39%). When the crude risk reduction in the subsequent time period was adjusted for the effects of differences in the risk for death, the 22% risk reduction declined to 18%. (Patients are classified according to treatment status at 12 months for follow-up after 1 year.)

For a more refined analysis of the effects of β-blockers after the first year, long-term survival curves are shown by treatment status at 12 months for each of the three clinical courses previously defined (Figure 1). Among the 1518 patients at lowest risk (stratum I), the mortality rate by the end of follow-up was close to 3.5% both in those on β-blocker therapy and in those off β-blocker therapy. In risk stratum II, although the mortality rate by 3 years after randomization was slightly greater in patients off β-blocker therapy (7.0%) compared with those on therapy (5.8%), the survival curves over time were virtually indistinguishable. In patients at highest risk, however, the overall survival experiences did differ significantly (P = 0.01 by log-rank test), and the continued use of β-blockers at 12 months was strongly associated with a reduced risk for death by the end of follow-up (11.9% compared with 20.8%).

Potential Sources of Bias

These data suggest that treatment is ineffective in patients at lower risk but may be effective for patients at highest risk based on characteristics of the clinical course during the first year after infarction. Before concluding that the findings are valid, certain potential sources of bias created by the use of postrandomization data must be considered.

Treatment Status Changes after Randomization

Treatment changes in the first year were often made in response to perceived changes in the patient's underlying clinical condition. If patients were taken off the assigned treatment (either propranolol or placebo) because of conditions that contraindicated the use of active therapy, the evaluation of comparative outcome rates would be less informative and perhaps misleading, because these patients would be ineligible for participation in a randomized, controlled trial and would be handled differently in actual clinical practice.

The potential effect of these withdrawals on later mortality comparisons was assessed by computing mortality rates after 1) removal of patients whose major reason for withdrawal from the BHAT treatment assignment was congestive heart failure and 2) after removal of all patients who were no longer on the originally assigned BHAT treatment. Although the data are not shown, the results of both sets of calculations were essentially the same as those presented.

The Use of Postrandomization Data To Define Risk Strata

In the consolidation process used to define the first-year clinical courses, the treatment status of all patients was ignored. If treatment affects the occurrence or intensity of clinical signs and symptoms used to define risk, however, it may in turn influence the classification of patients into risk strata. The potential for misclassification when stratifying patients by risk (a form of stage migration) is an unavoidable consequence of incorporating postrandomization data into an analysis of effectiveness. Interpretation of the differences in the mortality rates as measures of treatment effectiveness requires the assumption that the expression of these clinical features is not altered significantly by treatment.

The observations in Table 5 argue against the existence of this potential source of stage-migration-bias in the cohort of patients who survived to 12 months. If substantial stage migration had occurred, patients at higher risk would have been consistently shifted into lower risk strata because active treatment was significantly attenuating the expression of clinical features used in their definition. As shown, the severity and worsening of angina and congestive heart failure (events and dyspnea) during the first year of follow-up were, in fact, similar in patients originally assigned to propranolol compared with those originally assigned to placebo. Thus, it is unlikely that these findings were substantially affected by stage-migration bias.

Discussion

Methodologic Considerations

The inevitable heterogeneity in clinical courses that evolve from what appear, at the zero-time of a randomized, controlled trial, to be homogeneous populations (the treatment groups) implies that data on postrandomization dynamics must play a fundamental role in assessments of drug efficacy intended to guide long-term patient management. To accomplish such an assessment, we examined evidence from the largest and longest of the completed trials of β-adrenergic blocking drugs. These data represent an improvement over those traditionally available in observational studies, which often lack the methodologic rigor of a clinical trial and have limited opportunities for re-examining patients over time.

Given the nonrandomized nature of the surviving cohort after 1 year, however, analytic methods were needed to adjust for differences in underlying prognosis between treatment groups. One technique often used in prognostic studies that wish to adjust for characteristics that change over time is the Cox proportional-hazards model with time-dependent covariates. In such an analysis, the estimation of treatment effect is an average based on the accrual of patient time at risk under the varying treatment (and covariate) states. This approach was not considered suitable for our purposes because our aim was to examine the effect of β-blockade among patients who had all survived to 1 year.

There are alternative analytic techniques to protect internal validity that rely on more intuitive approaches to minimize susceptibility bias. These categorical techniques have a long history in clinical studies and include direct cross-stratification [18], automatic interaction detection [19], and asymmetric stratification [20]. By adopting a method of risk stratification broadly analogous to these categorical strategies, we have placed primary emphasis on the usefulness of the results for clinical practice.

The analysis we present may be viewed as complementary to those traditionally conducted for clinical trials, which, by their nature, rarely address the issue of duration of therapy. Although experimental studies may be designed to assess longitudinal changes in effectiveness by randomly stopping therapy after a period of time, the ethical constraints against withdrawing treatment when an overall benefit has been shown make alternative approaches to the assessment of continued benefit more attractive from a practical point of view.

When trials are analyzed based solely on a patient's status and treatment at the time of randomization, an implicit tradeoff is made about the type of information that is useful to patients and physicians. This tradeoff favors a pragmatic comparison of randomized patient groups over an explanatory comparison of treatments and clinical courses that may change during the conduct of the trial. The dilemma created by this tradeoff is that the analysis by randomized treatments provides a valid answer to only one of several important clinical questions (that is, the effectiveness of treatment as initiated). Other questions (for example, the effectiveness of treatment as received over time) are left unanswered.

Clinical Considerations

After 20 years of randomized trials, a clinical question remains about the length of time that patients should be maintained on β-blocker therapy after a myocardial infarction. Our analysis suggests that the estimation of net benefit for continuing to treat may be altered by a patient's prognosis as expressed by the clinical course that evolves after the acute event. As other investigators have suggested, the gain to those at low risk for death may not outweigh the medical and economic costs involved in such treatment [9, 12, 21, 22]. Our analysis confirms that in the setting of an uncomplicated first year after myocardial infarction, physicians and patients may want to reconsider the continued use of this medication. In contrast, for the 1-year survivors in this cohort who were categorized at highest mortality risk based on findings of recurrent ischemic events, cardiac arrhythmias, physical evidence of left ventricular dysfunction, or severe comorbidity during the first 12 months, the subsequent reduction in risk was estimated to be of the same magnitude as that observed overall for the first year.

Thus, data from the BHAT trial remain a major source of evidence supporting the use of β-blockers after a myocardial infarction, although this analysis has shown that, among 1-year survivors, continued treatment benefit appears to be restricted to patients at highest risk. The features used to define that risk are based on clinical characteristics readily available to all clinicians. Although continued treatment with β-blockers appears to be of limited benefit for patients judged to have a low-risk clinical course, it should be emphasized that the decision to maintain treatment in actual practice may be affected by considerations other than reduced mortality (for example, effects on morbidity or the presence of other indications for β-blockade, such as angina, hypertension, and supraventricular tachyarrhythmias).

In addition, the current findings are based on data from a randomized trial conducted before the development of treatment innovations over the past decade. Although the advent of a cointervention such as percutaneous transluminal coronary angioplasty or surgical revascularization may have altered the potential effect of β-blockade in reducing mortality, these interventions have not replaced the medical approach to secondary prevention, and β-blockers continue to be routinely prescribed for patients who have had a myocardial infarction. Although it is possible that these cotherapies might further affect the treatment benefit of β-blockers, we do not believe they would erode the importance of our analysis for high-risk patients.

The large-scale trials of β-blockade were designed to determine the benefit of a policy of initiating treatment in survivors of myocardial infarction. Our analysis has used data from one of these trials to address the issue of the effectiveness of long-term treatment. Potential problems with such alternative uses of trial data should not be ignored, but when a randomized design is not feasible for evaluating questions of clinical importance, such as longitudinal changes in drug efficacy, refinements and extensions of this type of observational method will, in our opinion, be necessary to guide patient management more effectively.