Predictors of Mortality and Mortality From Cardiac Causes in the Bypass Angioplasty Revascularization Investigation (BARI) Randomized Trial and Registry
Background—The impact of percutaneous transluminal coronary angioplasty (PTCA) and coronary artery bypass grafting (CABG) on long-term mortality rates in the presence of various demographic, clinical, and angiographic factors is uncertain in the population of patients suitable for both procedures.
Methods and Results—In the Bypass Angioplasty Revascularization Investigation (BARI) randomized trial and registry, 3610 patients who were eligible to receive PTCA and CABG were revascularized between 1989 and 1992. Multivariate Cox models were used to identify factors associated with 5-year mortality and cardiac mortality, with particular attention to factors that interact with treatment. Diabetic patients receiving insulin had higher mortality and cardiac mortality rates with PTCA compared with CABG (relative risk [RR] 1.78 and 2.63, respectively, P<0.001), and patients with ST elevation had higher cardiac mortality rates with CABG than with PTCA (RR 4.08, P<0.001). Factors most strongly associated with high overall mortality rates were insulin-treated diabetes, congestive heart failure, kidney failure, and older age. Black race was also associated with higher mortality rates (RR 1.49, P=0.019).
Conclusions—A set of variables was identified that could be used to help select a revascularization procedure and to evaluate risk of long-term mortality in the population of patients considering revascularization.
After 5 years of follow-up in the Bypass Angioplasty Revascularization Investigation (BARI) randomized trial, no statistically significant difference was observed in the primary outcome of mortality in patients with multivessel disease who were randomly assigned to a strategy of revascularization by PTCA versus CABG.1 However, a statistically significant advantage favoring CABG was observed for the secondary outcome of cardiac mortality,2 and after 7 years of follow-up, this advantage was also significant for overall mortality. At 5 and 7 years, the subgroup of randomized patients with treated diabetes was shown to have significantly better survival after CABG compared with PTCA, whereas the subgroup of nondiabetics had virtually identical survival after CABG and PTCA.1 3 The results from BARI and other similar trials4 5 6 suggest that a properly selected group of patients can be safely treated with either procedure; nonetheless, certain characteristics may identify better long-term outcome for an individual patient related to the selection of a particular revascularization strategy.
Population-based studies of revascularization have shown that the relation between specific clinical and angiographic variables and long-term mortality depend on the revascularization treatment received.7 8 9 However, the populations addressed in these studies were patients who received PTCA or CABG and not patients who were prospectively evaluated regarding eligibility for both therapies. In these studies, patients who received surgery generally had more severe angiographic profiles compared with patients who received angioplasty. This selection bias is a major limitation when comparing predictors of mortality within the 2 revascularization treatment groups.
The BARI trial completed the initial assessment of eligibility before approaching the patient for consent to randomization. The cohort of patients approached had homogeneous baseline characteristics and had been defined to be eligible to receive either CABG or PTCA. Patients who declined to consent to randomization were asked to consent to long-term follow-up, and if agreeable, were entered into a registry. The BARI randomized and registry population is uniquely suited to examine the predictive power of clinical and angiographic characteristics for 5-year survival among patients considering revascularization. The purpose of this investigation was to identify prognostically important clinical variables in this population and to determine whether the impact of these characteristics depends on the specific revascularization treatment received.
BARI was designed to compare an initial revascularization strategy with CABG versus PTCA. The protocol, patient characteristics, and clinical outcomes have been described in detail.1 2 3 10 Eligibility criteria were intended to identify a group of patients with clinically severe angina or documented ischemia requiring revascularization and coronary artery disease involving 2 or 3 vessels with characteristics amenable to revascularization by PTCA and CABG.
Initial screening identified 4107 eligible patients, of whom 1829 agreed to randomization. Of the 2278 patients who declined randomization, 2010 agreed to participate in the BARI registry, in which patients could choose their initial treatment with their physician but had follow-up similar to the randomized patients (Figure 1⇓). In the registry, “treatment” was defined as the first revascularization treatment received in the 3-month interval after study entry (196 patients did not receive any revascularization procedure in this time). The 1814 revascularized patients in the registry and the 1796 randomized patients who received their assigned treatment form the population analyzed in this report (Figure 1⇓).
Identical baseline data were collected for the randomized and registry patients. A central laboratory evaluated all ECGs. However, angiographic data used in these analyses were based on clinical site readings, since only clinical site interpretations were obtained for the registry patients.
Vital status was obtained for 98% of the patients as of June 5, 1995. An independent Mortality and Morbidity Classification Committee judged the cause of each death. The primary end point was death at 5 years, and cardiac death was a prespecified secondary end point. Average follow-up was 5.5 years.
An identical analysis strategy was used for the 2 end points: death and cardiac death. The variables presented in Table 1⇓ were designated as potential predictors. While constructing the multivariate models, missing values were replaced by the mean value for the individual variable except for ejection fraction (EF), in which 3 categories were used (normal, abnormal, and missing EF); thus, all patients were included in the multivariate analyses.
For each variable, preliminary multivariate Cox proportional hazards models were constructed that included the variable of interest, randomization group (randomized versus registry), treatment, and 2-way interaction terms between the variable and randomization group and the variable and treatment. All 2-way interactions with a level of P<0.10 were considered candidates for the final multivariate model. A multivariate Cox model including all main-effect variables was constructed, and the candidate interactions were allowed to enter the model in a stepwise fashion (entry criterion P<0.10). Then, a backward stepwise regression procedure was used for the main effects and the selected interactions (exit criterion P>0.10 for main effects and P>0.05 for interaction terms). Sex, race, and education were forced into the model at the last stage. Finally, models that did not adjust for randomization group were created separately to evaluate whether this variable was masking the influence of explanatory variables that were correlated with consenting to randomization.
Hazard ratios, 95% CIs, probability values, and normalized χ2 statistics were computed for the final multivariate Cox model. The normalized χ2 statistic is the Wald χ2 statistic for a given variable divided by the sum of the Wald χ2 statistics for all variables in the model. This statistic is analogous to the probability value, incorporating both the predictive power and the prevalence of the characteristic; however, it is scaled between 0 and 100, and the sum over all variables in the model equals 100.7 11
For each patient in the randomized trial, separate CABG-randomized, PTCA-randomized, CABG-registry, and PTCA-registry survival curves were estimated by means of the multivariate mortality Cox model and the patient-specific covariates. Then, for each treatment/randomization group, these individual patient estimates were averaged together to construct an adjusted survival estimate based on the BARI randomized patient cohort.12
Multivariate Predictors of Death
Table 3⇓ presents the final multivariate model for mortality. The only statistically significant interaction term was between treatment and insulin-treated diabetes (P=0.042). Compared with nondiabetic patients receiving CABG (ie, reference relative risk [RR] 1.00), nondiabetic PTCA patients had similar risk for death (RR 1.06), whereas diabetic CABG patients taking insulin had an ≈2-fold risk (RR 2.08) and diabetic PTCA patients taking insulin had a>3-fold risk (RR 3.70). The unadjusted Kaplan-Meier 5-year mortality rate estimates based on this interaction are provided in Table 4⇓.
Older age, kidney failure, congestive heart failure (CHF), left/mixed coronary dominance, abnormal/missing EF, diabetes treated with oral agents only, current smoker, peripheral vascular disease, hypertension, and history of malignancy were all powerful predictors (P<0.01) of higher mortality rates in the multivariate model. Black race was also a significant predictor (P=0.019) of higher mortality rates.
Adjusted Survival Estimates
The adjusted survival curves are presented in Figure 2⇓; adjusted 5-year estimates were 89.3% for CABG-randomized, 86.7% for PTCA-randomized, 91.1% for CABG-registry, and 90.5% for PTCA-registry.
Multivariate Predictors of Cardiac Mortality
Table 5⇓ presents the final multivariate model for cardiac mortality. In addition to the interaction between treatment and insulin-treated diabetes (P=0.030), the interactions between treatment and ST elevation (P=0.0064) and between treatment and class C lesions (P=0.021) were statistically significant. Unadjusted Kaplan-Meier curves were created for diabetic status and treatment (Figure 3⇓), and Kaplan-Meier estimates were computed for subgroups based on the combinations of insulin diabetes, ST elevation, and class C lesions (Table 6⇓).
In the multivariate model, insulin-treated diabetics had a higher risk of cardiac mortality compared with nondiabetics within the PTCA group (RR 2.90, P<0.001) and similar risk within the CABG group (RR 1.12, P=0.76). ST elevation was markedly associated with cardiac mortality within the CABG group (RR 4.14, P<0.001), whereas it had no measurable impact within the PTCA group (RR 1.00, P=1.0). The Kaplan-Meier estimates (Table 6⇑) indicate that regardless of the presence of class C lesions, there was a benefit associated with CABG for diabetic patients receiving insulin and a benefit associated with PTCA in patients with ST elevation; it was impossible to evaluate the group (n=15) that both received insulin and had ST elevation.
The increased cardiac mortality rate among the CABG patients with ST elevation was largely perioperative. In both the randomized and the registry groups, 22% of patients had a myocardial infarction (MI) <3 weeks before the initial procedure, and ST elevation was more common in patients who had a recent MI (9.3% versus 2.4%, P<0.001). With the use of identical multivariate cardiac mortality models, the interaction between ST elevation and treatment was significant in the subgroup of patients with a recent MI (P=0.0015) but not in the subgroup without a recent MI (P=0.50). Unadjusted Kaplan-Meier curves illustrate that the CABG patients with the combination of a recent MI and ST elevation had particularly high short-term cardiac mortality rate (16.1% at 2 months, Figure 4⇓).
Finally, in the multivariate model, the absence of class C lesions was associated with lower cardiac mortality rate in the CABG group (RR 0.57, P=0.021) but had little effect in the PTCA group (RR 1.14, P=0.46). The Kaplan-Meier estimates (Table 6⇑) suggest that among patients not receiving insulin and without ST elevation, there was a slight advantage favoring CABG for those without class C lesions, whereas the treatments were nearly equivalent for those with class C lesions.
Beyond the interaction terms, CHF, kidney failure, abnormal/missing EF, older age, diabetes treated with oral agents, ST depression, and left/mixed dominance were strong predictors of cardiac mortality (P<0.01). Although kidney failure was more prevalent in diabetics compared with nondiabetics (4.7% versus 1.6%, P=0.001), kidney failure was associated with a similar relative risk for cardiac mortality within the diabetic group and within the nondiabetic group. Lower education was marginally associated with higher cardiac mortality in this model (P=0.074) but was statistically significant when the randomization variable was excluded from the multivariate model (RR 0.85, P=0.049 for each level increase in education: grade school, high school, some college, college degree).
The BARI randomized and registry populations are ideal for investigating which factors influence long-term mortality for patients who are considering revascularization treatment alternatives. The present analyses suggest that among patients who have multivessel coronary artery disease and are eligible for both revascularization procedures, only a few evidence-based variables can be used to guide physicians in selecting a revascularization procedure for an individual patient.
Diabetic patients receiving insulin had a higher risk of death with PTCA compared with CABG, whereas other patients had a similar risk with PTCA versus CABG. The primary results from the BARI randomized trial showed that CABG was superior for patients with treated diabetes mellitus whether they received insulin or oral hypoglycemic drugs. In this study of the combined randomized and registry population, mortality rates over the 5 years of follow-up were similar with CABG and PTCA among diabetics taking oral hypoglycemic drugs but not taking insulin. Insulin-treated diabetes was the only variable found to have a significantly different impact regarding all-cause mortality in the 2 treatment groups.
Three factors were found to have differing effects on cardiac mortality, depending on the treatment received: insulin-treated diabetes, ST elevation, and presence of class C lesions. The interaction between ST elevation and treatment was highly significant despite the small number of patients with ST elevation at baseline (n=143). The result, which was that PTCA was superior to CABG for patients with ST elevation, was limited to the group of patients with a recent MI, and the excess mortality rates in the CABG group were seen primarily in the perioperative period. This is notable in the context of the recent VANQWISH (Veterans Affairs Non–Q-Wave Infarction Strategies in Hospital study) data, which showed a surprisingly high operative mortality rate in patients with non–Q-wave MI revascularized with CABG compared with conservative therapy but no such increased mortality rates with PTCA.13 Since ST-segment ECG data were not presented in the VANQUISH publication, these results may be entirely unrelated. In BARI, the presence of Q waves and ST depression were associated with worse long-term outcomes in both treatment groups.
The interaction between treatment and class C lesions was statistically significant, but the result, which was that the absence of class C lesions had a greater positive impact with CABG than with PTCA, goes against clinical intuition. The effect of class C lesions on the treatment comparison for all-cause mortality was not statistically significant. Moreover, patients with the most complex class C lesions were generally not suitable for PTCA and thus not included in BARI.
In other revascularization trials, Kaplan-Meier 5-year mortality rate estimates were 5.6% with CABG and 5.7% with PTCA in RITA-1 (Randomized Intervention Treatment of Angina), which randomized 1011 patients with single-vessel or multivessel disease,5 and 8.8% with CABG and 12.1% with PTCA in the Emory Angioplasty Versus Surgery Trial (EAST), which randomized 392 patients with multivessel disease from Emory University.4 Corresponding mortality rates in BARI were 10.5% for CABG-randomized, 13.6% for PTCA-randomized, 8.4% for CABG-registry, and 8.3% for PTCA-registry. The BARI registry patients, especially those receiving PTCA, had a lower risk profile than the randomized patients; consequently, the adjusted mortality rate estimates in the registry shifted toward the CABG-randomized rate, and this shift was more apparent in the PTCA-registry group.
Within the BARI population, the clinical history variables stand out as the major predictors of mortality. Classic cardiac risk factors, CHF, diabetes, hypertension, and kidney failure were powerful predictors of both cardiac and all-cause mortality. With the exception of age, the demographic/socioeconomic variables were less strongly associated with mortality rates. The randomized patients were significantly less educated than the registry patients, and lower education level was a significant independent predictor of cardiac mortality when the randomization variable was removed from the model. This implies that lower education was associated with cardiac mortality, and the randomization variable served as a surrogate for education in the model presented.
Black patients had a higher risk profile compared with nonblacks, including a larger proportion of women, lower education level, diabetes, and CHF. After adjusting for these differences, black patients had a significantly higher risk of death, and none of the interactions between sex, race, and education were statistically significant. Several studies have shown that black Americans are less likely than other Americans to receive coronary revascularization procedures,14 15 16 which may be partially explained by a greater reluctance to undergo invasive procedures.16 In past studies, however, survival did not differ by race among patients receiving PTCA17 or among Veterans Affairs cardiac patients (controlling for the use of invasive cardiac procedures).15 Black race was an independent predictor of not receiving an internal mammary artery graft in the BARI randomized surgery group,18 but black race remained a significant predictor of death after adjusting for internal mammary artery use during the initial procedure.
Poor left ventricular function and left or mixed coronary dominance were both powerful predictors of inferior outcome; otherwise, the angiographic factors had little influence on cardiac mortality. Since all BARI patients met specific angiographic eligibility criteria, the range of coronary disease represented in this group was rather narrow. In particular, patients with single-vessel disease and significant left main disease were excluded. Therefore, the BARI data may underemphasize the impact of angiographic characteristics regarding mortality in the entire population of patients undergoing revascularization. Nonetheless, it is notable that proximal left anterior descending coronary artery disease was not predictive of mortality in this population.
Although the BARI randomized clinical trial has adequate power for comparing CABG and PTCA, the power to detect small differences between the prognostic value of particular variables in the 2 treatment groups is somewhat limited. This population is ideal in the sense that it is the precise group in which the comparison between PTCA and CABG is relevant. However, the BARI population represents ≈12% of the patient cohort that undergoes revascularization,19 and as a result, the conclusions from this report are applicable only to the group of patients who are suitable both for PTCA and for CABG. Finally, since physicians chose treatments for patients in the BARI registry, it is conceivable that differential outcomes in the 2 treatment groups may have been influenced by unmeasured clinical signs or socioeconomic factors that were not included in this analysis.
This report identified numerous factors that were associated with mortality and cardiac mortality. Many of these characteristics were entirely expected, such as advanced age, history of CHF, and poor left ventricular function. It was unanticipated that few angiographic factors were influential and that black race was an independent predictor of worse survival.
The major purpose of this study was to identify factors that could be used to recommend one procedure over the other for an individual patient. First, CABG is preferable compared with PTCA for insulin-treated diabetics. Second, patients with ST elevation, particularly after a recent MI, may do better with PTCA compared with CABG. Finally, patients without class C lesions may have lower cardiac mortality rates with CABG rather than PTCA. Overall, these analyses suggest that a very restricted set of baseline factors can be used by the clinician to guide the selection of a revascularization strategy for patients with multivessel coronary artery disease. However, most patients who are deemed eligible for both procedures are likely to have similar long-term mortality rates with either CABG or PTCA as an initial procedure.
This study was supported by grants HL-38493, HL-38504, HL-38509, HL-38512, HL-38514-6, HL-38518, HL-38524-5, HL-38529, HL-38532, HL-38556, HL-38610, HL-38642, and HL-42145 from the National Heart, Lung, and Blood Institute, Bethesda, Md.
A complete list of the BARI Investigators has been published previously (Circulation. 1997;96:1761–1769).
- Received June 28, 1999.
- Revision received December 29, 1999.
- Accepted January 4, 2000.
- Copyright © 2000 by American Heart Association
Chaitman BR, Rosen AD, Williams DO, et al. Myocardial infarction and cardiac mortality in the Bypass Angioplasty Revascularization Investigation (BARI) randomized trial. Circulation. 1997;96:2162–2170.
BARI Investigators. Influence of diabetes on 5-year mortality and morbidity in a randomized trial comparing PTCA and CABG in patients with multivessel disease: the Bypass Angioplasty Revascularization Investigation (BARI). Circulation. 1997;96:1761–1769.
Barsness GW, Peterson ED, Ohman EM, et al. Relationship between diabetes mellitus and long-term survival after coronary bypass and angioplasty. Circulation. 1997;96:2551–2556.
Protocol for the Bypass Angioplasty Revascularization Investigation. Circulation. 1991;84(suppl V):V-1–V-27.
Lee KL, Pryor DB, Pieper KS, et al. Prognostic value of radionuclide angiography in medically treated patients with coronary artery disease: a comparison with clinical and catheterization variables. Circulation. 1990;82:1705–1717.