Coronary Revascularization in Diabetic Patients
A Comparison of the Randomized and Observational Components of the Bypass Angioplasty Revascularization Investigation (BARI)
Background—Patients with treated diabetes in the randomized-trial segment of the Bypass Angioplasty Revascularization Investigation (BARI) who were randomized to initial revascularization with PTCA had significantly worse 5-year survival than patients assigned to CABG. This treatment difference was not seen among diabetic patients eligible for BARI who opted to select their mode of revascularization. We hypothesized that differences in patient characteristics, assessed and unmeasured, together with the treatment selection in the registry, at least partially account for this discrepancy.
Methods and Results—Among diabetics taking insulin or oral hypoglycemic drugs at entry, angiographic and clinical presentations were comparable between randomized and registry patients. However, more registry patients were white, and registry diabetics tended to be more educated and more physically active and to report better quality of life. Procedural characteristics and in-hospital complications were comparable. The 5-year all-cause mortality rate was 34.5% in randomized diabetic patients assigned to PTCA versus 19.4% in CABG patients (P=0.0024; relative risk [RR]=1.87); corresponding cardiac mortality rates were 23.4% and 8.2%, respectively (P=0.0002; RR=3.10). The CABG benefit was more apparent among patients requiring insulin. In the registry, all-cause mortality was 14.4% for PTCA versus 14.9% for CABG (P=0.86, RR=1.10), with corresponding cardiac mortality rates of 7.5% and 6.0%, respectively (P=0.73; RR=1.07). These RRs in the registry increased to 1.29 and 1.41, respectively, after adjustment for all known differences between treatment groups.
Conclusions—BARI registry results are not inconsistent with the finding in the randomized trial that initial CABG is associated with better long-term survival than PTCA in treated diabetic patients with multivessel coronary disease suitable for either surgical or catheter-based revascularization.
The Bypass Angioplasty Revascularization Investigation (BARI) was designed to compare CABG and angioplasty (PTCA) in a randomized clinical trial.1 In addition, patients who were eligible for the BARI trial but were not randomized (and hence underwent CABG surgery or PTCA by choice rather than according to random assignment) were entered into an observational registry. Therefore, BARI provides an excellent opportunity to examine whether these 2 study mechanisms of evaluating the safety and efficacy of coronary revascularization in trial-eligible patients provide consistent conclusions. The case in point is the 5-year survival experience of treated diabetic patients in the randomized and registry cohorts of BARI.
As has been reported,2 overall 5-year survival, the primary end point of BARI, was not significantly different between the PTCA and CABG randomized-treatment arms. Moreover, there was no significant treatment difference within any subgroup of randomized patients, with the exception of treated diabetic patients, for whom 5-year mortality was highly significantly lower when the patients were initially treated with CABG (19%) than with PTCA (35%). This excess mortality in diabetic patients associated with PTCA was almost entirely attributable to cardiac death (15% 5-year cardiac mortality rate with initial PTCA versus 6% with initial CABG).3 Additionally, the substantial benefit of CABG was limited to the majority of patients who received internal mammary artery grafts during their index bypass surgery.
This important treatment effect found in the randomized trial (which had a probability of 3 in 1000 of being due to chance and which remained statistically significant after subgroup analysis was taken into account4 ) was not seen in the registry patients. Although the registry component did confirm that patients presenting with diabetes had an overall increased risk of 5-year mortality compared with nondiabetics and the magnitude of this relative risk (RR) was similar to that in randomized patients, the significant mortality difference between CABG and PTCA found in the randomized trial was not observed.3 In the present report, we present detailed information from our examination of the reasons for the discrepant findings between the 2 treatment comparisons of diabetic patients in BARI, the randomized and registry cohorts.
To be eligible for BARI, patients had to present with severe angina or ischemia and multivessel coronary artery disease suitable for initial revascularization both by PTCA and by CABG.1 Clinically eligible patients who consented to random assignment (n=1829) entered the randomized trial. An additional 2010 clinically eligible patients who did not consent to randomization but consented to provide follow-up data were entered into the registry. Among randomized patients, PTCA or CABG revascularization was performed within 2 weeks after study entry. In the BARI registry patients, initial treatment was defined as the first procedure that the patient received within 3 months of study entry.
Patients were considered to have treated diabetes mellitus if they received insulin or oral hypoglycemic drugs at the time of study entry. There were 353 such treated diabetic patients in the randomized trial (19%) and 339 in the registry (17%).
Of the registry diabetic patients, 117 were treated with CABG within 3 months of entry, 182 were treated with PTCA, and 38 underwent no revascularization procedures within 3 months and constituted the medically treated group. Treatment was undefined for 1 registry patient who died of cardiac causes 20 days after study entry and 1 who withdrew from the registry 6 days after study entry.
The BARI data collection methodology described in previous publications1 applies to the present report. Briefly, patients in the randomized trial were followed up with alternate annual clinic visits or telephone interviews, whereas registry patients were contacted only by telephone. Baseline angiograms for randomized patients were read locally at the clinical site and by a core radiographic laboratory. Because registry patients only had local clinical site readings of their angiograms, clinical site readings were used for the comparison of angiographic characteristics in the present report. Significant lesions were defined as those with stenosis ≥50% in a vessel >1.5 mm in diameter.
Study End Points
The primary outcome of BARI was mortality from all causes. The vital status of each patient as of June 5, 1995, was determined. The primary cause of death was categorized by the Morbidity and Mortality Classification Committee. Specific classifications for cause of death included cardiac, noncardiac but atherosclerotic, other medical cause, and nonmedical (eg, accident or suicide). Cardiac cause of death included direct causes, such as cardiogenic shock, and contributory causes, such as chronic congestive heart failure with terminal pulmonary embolism.
Differences in baseline characteristics, as well as in-hospital outcomes between randomized versus registry patients and between treatment groups within the registry, were assessed by χ2 tests or Fisher’s exact tests for dichotomous variables and t tests or Wilcoxon tests for continuous variables.
Kaplan-Meier estimates for crude total and cardiac mortality rates were computed and compared (by treatment) with the log-rank test.5 Multivariate Cox regression analysis6 was used to assess the risk of death associated with treatment, with adjustment for other risk factors. Registry patients treated medically or with undefined treatment were excluded from multivariate analyses. In the construction of multivariate models, variables associated with mortality were incorporated into the Cox regression models for total mortality and cardiac mortality. In addition, variables for which the distribution differed between randomized and registry diabetic patients, as well as those that differed between PTCA and CABG treatment groups in the registry patients, were included in the multivariate models. Owing to the limited number of patients, we did not consider variables with substantial missing values for patients with events, notably left ventricular ejection fraction. Finally, in an attempt to adjust for the many unmeasured clinical and demographic factors that may relate to mortality, self-reported quality of life was included in all Cox models as an index variable for the overall health condition of the patients. The 11% of registry patients with missing self-reported quality-of-life assessments were treated as a separate category in this analysis.
Randomized Versus Registry Patients
As shown in Table 1⇓, registry diabetic patients were more likely to be white, were more educated, were less likely to smoke, were more physically active, and reported better self-evaluated quality of life than randomized diabetic patients. Congestive heart failure was reported less frequently by registry patients. Fewer registry diabetic patients required insulin treatment than in the randomized study (39% versus 46%; P=0.065). Other demographic and clinical characteristics, including angina pattern and ECG characteristics, were similar for randomized and registry patients. Angiographic profiles were also generally similar, with the exception of somewhat better left ventricular ejection fraction in the registry patients.
CABG Versus PTCA in Registry Patients
Within the registry diabetic patients, the CABG and PTCA intervention groups were significantly different with respect to several angiographic characteristics. Registry patients treated with PTCA had lower prevalences of triple-vessel disease (P<0.001), ≥4 significant lesions (P<0.05), and significant proximal left anterior descending coronary artery lesions (P<0.001), as well as fewer ostial lesions (P<0.05). The registry PTCA patients also had the lowest prevalence of insulin use at baseline.
Procedural Characteristics and In-Hospital Outcome
Table 2⇓ illustrates the initial procedural characteristics and in-hospital outcome of randomized and registry CABG patients. The CABG intervention groups had 3.5 and 3.8 mean significant lesions, respectively, for randomized and registry patients, and the mean number of grafts received per patient was 2.9 and 3.0, respectively. The surgical procedure involved ≥1 internal mammary artery graft in 81% of randomized and 87% of registry patients (P=0.15), but significantly fewer registry patients had sequential grafts. In registry patients, 96% of all intended vessels were grafted, compared with 87% among randomized patients (P=0.015). More registry patients required post-CABG inotropic support.
There was no statistically significant difference in the occurrence of in-hospital death, Q-wave myocardial infarction, or stroke between randomized and registry diabetic patients receiving CABG, although all of these events occurred more frequently in the registry patients. The incidence of wound dehiscence or infection was significantly higher in registry patients (7.7% versus 1.7% in randomized patients; P=0.016).
Table 2⇑ also shows the procedural characteristics and in-hospital events in diabetic patients treated with PTCA. The average number of significant lesions was 3.4 for randomized and 3.3 for registry patients, and the mean number of significant lesions attempted was 2.3 and 2.3, respectively. Randomized and registry patients had similar procedural angiographic success rates (76% and 74% of all attempted significant lesions successfully dilated, respectively). However, mean postprocedural stenosis was significantly lower in registry patients (26.3% versus 29.4% in randomized patients; P=0.019). In addition, 2% of registry patients were unstable when they left the catheterization laboratory compared with 6% of the randomized patients (P=0.047). Major in-hospital complications occurred only in randomized patients. Abrupt closure rates were comparable, whereas the rate of emergency CABG was higher in randomized patients (7.1% versus 3.3% in registry).
Figure 1⇓ shows cumulative rates of survival and freedom from cardiac death for the 4 subgroups of patients receiving intervention. Randomized diabetic patients undergoing PTCA had significantly higher total 5-year mortality rates (Kaplan-Meier rates of 34.5% versus 19.4%; P=0.0024 by log-rank test) and higher 5-year cardiac mortality rates (23.4% versus 8.2%; P=0.0002) than randomized patients treated with CABG. As seen in the survival curves, the discrepancy in cumulative survival by initial treatment increased with time for the randomized patients. In the registry population, however, PTCA- and CABG-treated diabetic patients had similar cardiac mortality rates (7.5% and 6.0%; P=0.73) and total mortality rates (14.4% and 14.9%; P=0.86). In each cohort, approximately half of all deaths were due to direct or contributory cardiac causes. For both randomized and registry patients, rates of death due to noncardiac causes were similar in the PTCA and CABG groups. We note that as a whole, randomized patients had higher 5-year mortality rates than registry patients.
Figures 2⇓ and 3⇓ present total and cardiac survival curves for randomized and registry diabetic patients classified according to use of insulin at baseline (versus use of oral hypoglycemic agents only). In randomized patients, the treatment advantage of CABG was evident for both types of diabetic patients. It is apparent that diabetics using insulin who were randomized to PTCA had an excess of events occurring soon after the initial intervention, whereas the disadvantage of PTCA compared with CABG for diabetics using oral hypoglycemic agents became apparent only several years after the intervention. In the registry patients, there was only a trend of long-term increased cardiac mortality for insulin-requiring diabetics treated with PTCA.
As shown in Table 3⇓, the unadjusted RR of 5-year total and cardiac mortality was significantly higher than 1.0 (RR=1.87 for all-cause mortality and 3.10 for cardiac death) for PTCA compared with CABG in the randomized study population with treated diabetes. A treatment effect for all-cause mortality was not found in the diabetic registry patients. The magnitude of the RR in this registry population increased only slightly (from 1.10 to 1.26) after adjustment for clinical baseline risk factors including congestive heart failure, age, and insulin use at baseline. Further adjustment for angiographic risk factors (extent of vessel disease, number of significant lesions, and presence of significant proximal left anterior descending coronary artery lesions) and self-reported quality of life did not further affect the adjusted RR for all-cause mortality.
A substantial effect of statistical adjustment, however, is seen in the registry patients for the outcome of cardiac mortality. In registry diabetic patients, this RR increased from no excess risk with PTCA in the crude comparison (unadjusted RR=1.07) to 1.23 after adjustment for clinical risk factors alone, increased further to 1.35 after additional adjustment for angiographic variables, and finally increased to a 41% higher risk (RR=1.41, although still not statistically significant) after controlling for self-reported quality of life along with these other factors.
This report provides several plausible explanations for the apparent inconsistency regarding treatment effect in randomized versus registry diabetic patients in BARI. Diabetic patients in the registry were similar to those consenting to randomization with respect to clinical and angiographic variables, with the exception of history of congestive heart failure, the prevalence of which was slightly lower in the registry, and ejection fraction, which was slightly higher. The numerous significant differences between patients consenting and not consenting to randomization were limited to variables not related to BARI eligibility. These included race, education, smoking, physical activity level, and self-reported quality of life. It is possible that these factors, some of which have been shown to influence adherence to prescribed medications in other studies,7 may influence compliance with treatment for both diabetes and coronary artery disease in BARI. Also, because BARI was not designed to focus on diabetic patients, the severity and duration of diabetes, which were not assessed, could have differed between the treatment groups. Because survival with PTCA (but not with initial CABG) was dependent on insulin use at baseline in both the randomized and registry patients, it is likely that other measures related to diabetes control would also affect this outcome.
Although balanced distribution of unmeasured variables between treatment groups in the trial is assured by the randomization mechanism, we would expect these factors to be substantially imbalanced between treatment groups in the registry. In particular, the choice of initial procedure in this setting is complex and highly dependent on each patient’s individual preference and the physician’s judgment. The choice of initial treatment in the registry, as seen in our detailed examination of the factors that were measured, was definitely guided by the extent of coronary disease. After either consent to randomization or treatment selection, the registry PTCA patients ended up with the fewest clinical and socioeconomic risk factors, the registry CABG patients had the worst angiographic profile, and the randomized patients were “in the middle” angiographically with a relatively unfavorable socioeconomic profile. Therefore, direct comparisons of treatment arms within the registry, as well as crude comparisons of randomized versus registry patients receiving the same treatment, are substantially biased. To the limited extent that controlling for imbalances by treatment between the registry diabetic patients was possible, sequential adjustment for clinical factors, angiographic factors, and quality of life shifted the estimate of treatment effect on cardiac survival somewhat in favor of CABG. In this sense, the registry analysis is consistent with the findings of the randomized trial.
The difference between the treatment effect in the randomized and registry patients, although certainly not negligible, should not influence the conclusions of the randomized trial. The BARI trial demonstrated unequivocally that for selected diabetic patients with multivessel coronary disease, CABG is a superior initial treatment strategy. The differences in demographic presentation as well as treatment effect between the randomized and registry population, however, illustrate the possibility that the strong advantage of CABG seen in the randomized population may not extend to all diabetic patients eligible for BARI. At the same time, we cannot rule out that CABG is advantageous for all BARI-eligible diabetic patients.
Given the various sources of bias and confounding, both apparent and subtle, once we step just slightly outside of the bounds of the BARI randomized trial, we are not surprised at the inconsistent findings of other observational studies (as well as clinical trials) that attempt to refute or validate the BARI randomized trial results or that attempt to compare the 2 revascularization approaches in more wide-ranging populations of diabetic patients. In addition to sharing the inability of our analysis to directly assess diabetic self-management, the patient selection criteria in these studies often only roughly approximate those used in BARI. For example, Barsness et al8 attempted to simulate the BARI study in a retrospectively selected population of revascularized patients in the Duke database. Although all patients who did not meet BARI clinical eligibility criteria were excluded from the analysis, it is not possible on the basis of retrospective review of general angiographic criteria to definitively determine whether patients who underwent CABG could also have undergone PTCA. It is not possible to simulate the BARI setting, in which determination of PTCA eligibility was based on operator review of the actual angiogram and implied a 50% likelihood of having to actually perform the procedure, in a retrospective database study. In the BARI trial, this requirement eliminated two thirds of otherwise eligible patients.
In a second retrospective study, Weintraub et al9 analyzed 2639 diabetic patients with multivessel disease undergoing revascularization between 1981 and 1994. All such revascularized patients treated outside the setting of acute myocardial infarction were included, and as in the BARI registry, patients treated with CABG tended to have more severe angiographic disease. After adjustment for baseline differences, they found that, as in BARI, insulin-requiring diabetic patients treated with PTCA fared significantly worse than those treated with CABG in terms of long-term survival. The hazard ratio of 1.35 was of lower magnitude than in the BARI randomized trial but was similar in magnitude to the adjusted ratios reported here for the BARI registry (1.29 for all-cause mortality and 1.41 for cardiac mortality).
We have seen in our analysis how a randomized population can differ in important ways from the entire study-eligible population that the trial was designed to address. Such a difference may also play a part in the inconsistent findings of treatment effect in other randomized diabetic populations, although more explicit differences in eligibility criteria and small sample sizes no doubt play a role. Specifically, the Emory Angioplasty Versus Surgery Trial (EAST) reported no benefit of CABG in diabetic patients at 3 years,10 whereas a benefit of CABG was found in the Coronary Artery Bypass Revascularization Investigation (CABRI).11
In conclusion, although the BARI randomized and registry diabetic patient populations are clinically quite similar, these 2 groups of patients are different in many important respects. These differences, together with the sizable angiographic differences between the registry patients who selected CABG and those who selected PTCA, preclude an expectation of identical long-term event rates and identical treatment effects. Moreover, the lack of information on the extent and control of diabetes limits the power of our adjusted analyses to elucidate the true registry treatment effect. Nevertheless, there are 2 findings that lend a degree of support to the randomized study findings previously presented. First, the finding of a stronger treatment effect among randomized patients using insulin at study entry, which was also seen to a smaller extent in the registry patients, suggests a dose-response treatment difference between CABG and PTCA that increases with severity of diabetes. Second, the adjusted treatment difference in the registry patients, although not statistically significant, does point to a possible advantage of initial CABG in these patients as well. As already discussed, we must leave open the possibility that in certain diabetic patients with coronary disease, the physician can, as was done in the registry, choose a suitable revascularization approach based on clinical sense alone. Evidence-based criteria for this selection remain to be developed for the many diabetic patients with coronary disease.
- Received July 28, 1998.
- Revision received October 5, 1998.
- Accepted October 22, 1998.
- Copyright © 1999 by American Heart Association
Protocol for the bypass angioplasty revascularization investigation. Circulation. 1991;84(suppl V):V1–V27.
The 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.
Brooks MM, Rosen AD, Holubkov R, Kelsey SF, Detre K. Treatment comparisons controlling for multiple testing. Control Clin Trials. 1997;18:81S. Abstract.
Cox DR. Regression models and life-tables. J R Stat Soc (B). 1972;34:187–220.
Schron EB, Brooks MM, Gorkin L, Kellen JC, Morris M, Campion J, Shumaker SA, Corum J. Relation of sociodemographic, clinical, and quality of life variables to adherence in the cardiac arrhythmia suppression trial. Cardiovasc Nurs. 1996;32:1–6.
Barsness GW, Peterson ED, Ohman EM, Nelson CL, DeLong ER, Reves JG, Smith PK, Anderson RD, Jones RH, Mark DB, Califf RM. Relationship between diabetes mellitus and long-term survival after coronary bypass and angioplasty. Circulation. 1997;96:2551–2556.
CABRI. Long-term follow-up of European revascularization trial. Presented at the 68th Scientific Sessions, Plenary Session XII, of the American Heart Association; November 16, 1995; Anaheim, Calif.