Survival difference between coronary bypass surgery and percutaneous coronary intervention

Abstract

Objectives. To assess whether the previously observed lower death rate with coronary artery bypass surgery compared with percutaneous coronary intervention in subsets of patients with coronary artery disease persists in more recent years. Design. Retrospective study from Feiring Heart Clinic database of survival in 17739 patients followed for 5 years after coronary revascularization. The cohorts treated in 1999–2005 and 2006–2011 were compared using Cox regression and propensity score analyses. Results. Cox regression and propensity score analyses revealed no difference in survival in either time period in one- and two-vessel diseases. In three-vessel disease, the hazard ratios between bypass surgery and percutaneous intervention were 0.62 (95% confidence interval [CI]: 0.53–0.71, p <0.001) and 0.59 (95% CI: 0.47–0.73, p < 0.001), respectively, in the two time periods, indicating persistent higher survival with bypass surgery. Conclusions. The previously observed lower death rate of coronary artery bypass surgery compared with percutaneous intervention in patients with three-vessel disease is persistent in more recent years and indicates that bypass surgery still should be the standard treatment for these patients.

Key words::

• coronary bypass surgery
• percutaneous coronary intervention
• survival

Introduction

The treatment of coronary artery disease is continuously evolving as the treatment options change with innovations in drugs and equipment for coronary artery bypass grafting (CABG) and percutaneous coronary interventions (PCIs). Thus, the selection of treatment options has been and still is a matter of debate. Traditionally, PCI has been selected for the simpler cases and CABG for coronary disease with more complex pathology. A number of reports have been published seeking to provide guidelines for the selection process in the light of subsequent survival and clinical events in different subsets of coronary anatomy (1–8). Especially, there has been a focus on the impact of drug-eluting stents (DESs) on subsequent survival. The results have, however, been conflicting (9–12).

In 2007, we published the results from the database of Feiring Heart Clinic of five-year survival after PCI or CABG adjusting for baseline variables. The conclusions from that study were a higher survival rate for CABG-treated patients with three-vessel disease compared with patients treated with PCI (2). The aim of the present study was to investigate whether this higher survival rate still is present in the subsequent years after the first report.

Materials

Feiring Heart Clinic has had a common database for cardiological and surgical department since 1999. This database contains information on demographics, clinical and angiographic parameters, treatment, diagnosis (ICD 10), and surgical operative codes (Nordic Medico-Statistical Committee Classification of Surgical procedures). Only patients submitted to angiography and the subsequent treatment at our institution were included in the analyses. The survival status as of March 22, 2012 was established through the Norwegian National Registry, which also gave formal consent to obtain the data. The endpoint was death by any cause. By the end of 2011, a total of 17739 patients had been treated with either PCI (12418 patients) or CABG (5321 patients). Patients with a combined operation of valves and bypass surgery were excluded from these analyses. Emigrated patients were censored at the date of emigration. The patients were allocated to the treatment groups according to the strategy chosen at the first admittance to the institution and only information from that admittance was used in the analyses. Thus, each patient could enter the study only once. The use of DES was relatively rare in the institution before 2006, but rose sharply after that (12). Other aspects from this database have previously been published (13–15).

Statistical analyses

The purpose of the analyses was to evaluate whether the difference in survival rate of patients treated with PCI or CABG previously observed was influenced by the time of the initial treatment. Our data on this difference from patients treated from 1999 to 2006 have previously been reported (2). We therefore decided to compare the survival rate between PCI- and CABG-treated patients in two different time periods (1999–2005 and 2006–2011), using the same method and multivariable models as previously reported (2). The data were analyzed on an intention-to-treat basis.

Continuous covariates were tested with the skewness and kurtosis test for normality, and were evaluated by the Kruskal–Wallis test if they were deviating from normal distribution, if not analysis of variance was used for testing difference between cohorts. Categorical covariates were assessed by Fisher's exact test or chi-square test in the case of excessive permutations. Univariate analyses of survival were performed using the Kaplan–Meier product limit estimator and the log-rank test. In the multivariable analyses, missing values for covariates were substituted by imputation using best subset multiple regression based on the other covariates. Multivariable survival analyses adjusting for baseline differences were performed using a Cox proportional hazard model. The basic model for the analyses was the final model obtained in the previous report (2). Baseline variables were tested and added to this model by manual forward selection process. A categorical variable combining treatment choice and time period with four levels was forced into the final model. In the multivariable model, continuous variables were tested for linearity in ln hazard by quartile plots. The assumption of proportional hazard was checked by a test based on Schoenfeld residuals and by log–log plots. Interactions were kept in the model if they were statistically significant and biologically interesting. The final model was evaluated with the linktest and by Grønnesby–Borgan test for goodness of fit. Selection bias was addressed by propensity score analyses. Baseline covariates were fitted in a logistics regression model predicting treatment allocation (PCI = 0 and CABG = 1). Significant covariates and interactions were kept in the model, which were subsequently checked with Hosmer–Lemeshow test for goodness of fit and c-statistics [area under the receiver operating characteristics (ROC) curve]. The propensity score was calculated from a probit analysis using the model obtained from the logistic regression analysis. The propensity score was used as a single covariate in Cox regression and as a matching variable in 1:1 matching without replacement and a caliper of < 0.01 for difference in propensity score. The matched pairs were then used in a stratified Cox regression analysis. All analyses were done using STATA version 12 (College Station, Texas, USA) and the propensity score matching with the STATA program psmatch2.

Results

From the database, a total of 17739 patients were identified from their first admittance to the institution with known survival status obtained from the Norwegian National Registry as of March 22, 2012. The endpoint was mortality from any cause after 5 years. There were a total of 1653 deaths and a total of 73475 patient-years at risk. The patients were divided into four groups based on the time period for the initial treatment (1999–2005 and 2006–2011) and treatment allocation (PCI or CABG). Baseline demographics, and clinical and angiographic variables in the four groups are depicted in Table I.

Table I. Baseline demographic and clinical variables.

Variable	1999–2005 PCI N = 6749	2006–2011 PCI N = 5669	1999–2005 CABG N = 3926	2006–2011 CABG N = 1395	P value
Age years (mean ± SD)	63.8 ± 11.0	66.0 ± 10.9	66.9 ± 9.9	67.1 ± 9.6	< 0.001
Gender (male/female) %	73.3/26.7	71.8/28.2	77.5/22.5	77.0/23.0	< 0.001
Ejection fraction % (mean ± SD)	67.7 ± 13.2	67.6 ± 11.4	67.3 ± 13.7	67.8 ± 11.8	0.6
LVEDP mmHg (mean ± SD)	13.4 ± 4.9	15.0 ± 4.7	14.2 ± 5.6	16.2 ± 5.5	< 0.001
Generalized arteriosclerosis %	2.8	5.7	3.9	1.4	< 0.001
Other significant disease %	2.6	8.7	2.7	6.2	< 0.001
Ischemia on EECG %	32.7	19.7	48.5	32.8	< 0.001
CCS function class %
0	2.1	0.9	1.9	1.3
I	7.9	3.9	5.3	4.0
II	34.3	34.4	31.3	26.8
III	38.0	46.3	46.2	52.3
IV	17.7	14.6	15.3	15.7	< 0.001
Unstable angina %	26.0	29.5	25.9	30.2	< 0.001
Diabetes %	11.5	14.2	14.8	16.0	< 0.001
Hypertension %	28.3	33.5	34.4	34.1	< 0.001
Current smoker %	25.9	16.5	22.1	14.4	< 0.001
Previous myocardial infarction %	48.4	28.0	47.8	28.5	< 0.001
Previous PCI %	15.0	18.9	6.6	10.5	< 0.001
Previous CABG %	14.1	13.2	2.7	1.7	< 0.001
Radial entry site %	85.1	78.8	85.4	61.5	< 0.001
Coronary angiography
One-vessel disease %	42.6	48.8	1.9	2.0
Two-vessel disease %	30.8	29.5	15.7	18.5
Three-vessel disease %	26.6	21.7	82.4	79.5	< 0.001

LVEDP, left ventricular end diastolic pressure. Generalized arteriosclerosis is defined as “previous extra-cardiac arteriosclerotic symptoms.” EECG, exercise ECG. CCS, Canadian Cardiovascular Society functional class for angina. Diabetes is defined as “previously known and treated diabetes with diet or drugs.” Other significant disease is defined as “renal, hepatic, or pulmonary disease and serious obesity deemed of importance in the treatment at the discretion of the physician.” Unstable angina also includes patients with non-ST-elevation myocardial infarction.

A Kaplan–Meier plot of the mortality in the four groups is shown in Figure 1. The log-rank test revealed a highly significant difference between the four groups (p < 0.001).

Figure 1. Kaplan–Meier estimate of mortality with significant difference between the four groups (p < 0.001—log-rank test).

As can be seen from Table I, there are considerable baseline differences between the groups and time periods that are expected to affect mortality. In the last time period, there is a considerable decline in percent smokers, patients with previous myocardial infarction, and detection of ischemia on exercise ECG, while the prevalence of diabetes has increased. The variables in Table I were tested for inclusion in the model and the final model contained 10 main effects and one interaction in addition to the categorical time period/treatment allocation variable (Table II). The interaction between number of diseased vessels and time/treatment variable was highly significant, definitely biologically interesting and was also found in the analyses in 2007. The results are therefore given separately for one- two- and three-vessel diseases. The global test for proportional hazard based on Schoenfeld residuals was negative (p = 0.08) for the whole time period from 1999 to 2011, and also tested separately for each period (1999–2005 and 2006–2011, p = 0.46 and 0.78, respectively). Based on this test, and evaluation with plots of Schoenfeld residuals over time for individual covariates of borderline significance in proportional hazard assumption and log–log plots, it was decided to keep all the covariates in the model unchanged. The final model was tested with the linktest which was negative (p = 0.51) and so was the Grønnesby–Borgan test for goodness of fit (p = 0.39).

Table II. The complete multivariable model.

Variable	Hazard ratio	95% CI	p value
Age (/10 year)*	2.26	2.14–2.41	< 0.001
Ejection fraction (/5%)*	0.87	0.85–0.88	< 0.001
Gender	1.23	1.09–1.38	< 0.001
Smoking	1.39	1.22–1.57	< 0.001
Diabetes	1.39	1.22–1.57	< 0.001
Ischemia on exercise ECG	0.71	0.63–0.81	< 0.001
CCS function class	1.25	1.17–1.34	< 0.001
Number of diseased coronary arteries	1.38	1.26–1.51	< 0.001
Generalized arteriosclerosis	1.26	1.03–1.53	0.022
Other significant disease	2.57	2.17–3.04	< 0.001
Interaction number of diseased vessels and strategy/epoch	0.85	0.77–0.93	0.001
PCI 1999–2005	Reference value = 1
PCI 2006–2011	1.05	0.89–1.24	0.565
CABG 1999–2005	1.20	0.85–1.70	0.294
CABG 2006–2011	1.20	0.71–2.05	0.50

*The hazard ratios are calculated for each 10-year increment in age and 5% increment in ejection fraction, respectively.

Disregarding treatment strategy and only testing the two eras (1999–2005 and 2006–2011), the five-year survival was significantly better in the most recent time period [hazards ratio (HR): 0.86, 95% confidence interval (CI): 0.77–0.96, p = 0.009]. Subgroup analyses revealed that this effect mainly resided in patients with three-vessel disease (HR: 0.80, 95% CI: 0.69–0.93, p = 0.004) and was present in both treatment strategies.

In Table III, the HRs for CABG versus PCI in the two time periods are depicted for one-, two-, and three-vessel diseases. When comparing the results of survival between the two treatment strategies in the two time periods, the survival in one- and two-vessel diseases is not significantly different in any time period. In three-vessel disease, however, the survival of CABG patients is significantly better in both time periods with HR of 0.62 and 0.59, respectively. The lower death rate of CABG- compared with PCI-treated patients with three-vessel disease was not different in the two time periods (HR = 0.95, 95% CI: 0.73–1.22, p = 0.70).

Table III. Comparison across strategy/time periods and number of diseased vessels calculated from the multivariable Cox model.

Strategy/time period	One-vessel disease		Two-vessel disease		Three-vessel disease
	HR (95% CI)	p value	HR (95% CI)	p value	HR (95% CI)	p value
PCI 1999–2005	Reference cohort
CABG 1999–2005	1.20 (0.85–1.20)	0.294	0.86 (0.72–1.03)	0.110	0.62 (0.53–0.71)	< 0.001
PCI 2006–2011	Reference cohort
CABG 2006–2011	1.14 (0.74–1.77)	0.553	0.82 (0.61–1.06)	0.179	0.59 (0.47–0.73)	< 0.001

Alternative Cox modeling with separate models for one-, two-, and three-vessel diseases resulted in too few events for meaningful analysis in one-vessel disease. Modeling one- and two-vessel diseases together and three-vessel disease separately yielded practically identical results as in the model with interaction (not shown).

The final model for the estimation of propensity score contained 11 significant main effects and 8 interactions. The Hosmer–Lemeshow test for goodness of fit was negative (p = 0.68), and the c-statistics (area under ROC curve) was 0.90. The propensity score was then calculated from this model. The propensity score for the PCI cohort was [mean ± standard deviation (SD)] 0.17 ± 0.23 (range: 0.0004–0.93), and for the CABG cohort it was 0.63 ± 0.24 (range: 0.005–0.95). Since the general Cox model indicated different treatment effects according to number of diseased vessels, the analyses with propensity score were conducted separately for one-, two-, and three-vessel diseases (Table IV). No significant difference was found in one- and two-vessel diseases in either period. In three-vessel disease, there was a significantly higher survival rate in the CABG groups in both periods compared with that in PCI-treated patients, with a slightly higher rate of relative survival in the last time period (HR = 0.72, 95% CI: 0.53–0.98, p = 0.039) compared with the difference in the first period.

Table IV. Comparison across strategy/time periods and number of diseased vessels calculated from the Cox model with propensity score as single covariate.

Strategy/time period	One-vessel disease (N = 5165)		Two-vessel disease (N = 4285)		Three-vessel disease (N = 7011)
	HR (95% CI)	p value	HR (95% CI)	p value	HR (95% CI)	p value
PCI 1999–2005	Reference cohort
CABG 1999–2005	1.61 (0.71–3.63)	0.254	1.00 (0.72–1.37)	0.979	0.64 (0.54–0.76)	< 0.001
PCI 2006–2011	Reference cohort
CABG 2006–2011	0.53 (0.07–3.79)	0.526	0.57 (0.32–1.02)	0.059	0.46 (0.35–0.61)	< 0.001

The propensity scores were also used in matching analyses. The matching was done with a caliper set to 0.01 in the propensity score. The overall standardized % bias in the covariates was reduced from (mean ±SD) 19.8 ± 35.4 to 4.2 ± 3.2 by the matching process. There were relatively few matched pairs in one-vessel disease, and the analyses were therefore done in one- and two-vessel diseases together and three-vessel disease separately. The matched pairs were used as a stratifying variable in subsequent Cox regression. In one- and two-vessel diseases, there were 946 matched pairs for analysis resulting in a reduction in the standardized percent bias from (mean ± SD) 14.8 ± 27.9% to 2.8 ± 2.0%. The stratified Cox analyses revealed no significant difference in survival in any of the combinations of time period and treatment strategies. For three-vessel disease, there were 1601 pairs for comparison. The reduction in standardized bias was from (mean ± SD) 18.6 ± 26.9% to 8.4 ± 10.7%. There was a markedly higher survival rate for CABG-treated compared with PCI-treated patients in both time periods with HR of 0.56 (0.24–0.64, p < 0.001) in the first period and 0.31 (0.17–0.54, p < 0.001) in the second. A comparison of the differences in the two periods indicates no difference in the higher survival rate of CABG-treated compared with PCI-treated patients in the two periods (HR: 0.55, 95% CI: 0.27–1.09, p = 0.088).

Discussion

The aim of the present study was to assess the survival rates after treatment with PCI or CABG in subgroups of patients with coronary artery disease by comparing two time periods of initial treatment (1999–2005 and 2006–2011). The basis for this comparison was to adopt the models used in 2007 (2) to the additional data acquired in the subsequent years. In the original report, patients were included in the analyses if they were treated before December 31, 2005. It was therefore reasonable to construct the time periods as before and after that date. The length of follow-up was set to 5 years as a compromise between long-term follow-up and a reasonable number of patients still at risk at the end of the observation period. This was set prior to the analyses of the data.

The survival was analyzed with three separate methods: ordinary multivariable Cox analyses, Cox regression based on propensity scores, and propensity score matching analyses. The multivariable analyses with the regular Cox model indicated a strong interaction between treatment strategy and number of diseased vessel, just as in the analyses in 2007 (2). All results were therefore reported in the subgroups of number of diseased vessels. The three methods yielded identical results with no differences in survival in patients with one- and two-vessel diseases. In three-vessel disease, the analyses revealed consistent results with a lower death rate in patients treated with CABG. None of the methods indicated that this difference was smaller in the most recent time period. In the regular Cox analyses and propensity score matching, the higher survival of CABG-treated compared with PCI-treated patients was not different in the two time periods, while usage of the propensity score as a single variable in the Cox model indicated that the difference was actually slightly larger in the most recent time period. The increase, however, was of borderline significance.

The discussion of the best treatment strategy in patients with coronary artery disease has been going on for years. Until recently, several larger studies addressing this problem have been observational studies (1–5). The general pattern emerging from these studies has been a higher survival rate for patients with complex coronary disease treated with CABG. Although an impressive number of studies including a large number of patients have contributed to this observation, selection bias cannot be completely ruled out. The early randomized studies were in general too small to detect a difference in mortality, but a meta-analysis by Hoffman et al. (6) largely supported the results from the observational studies. This analysis was published in 2003, and patients were included in the studies from 1987 to 1999, probably indicating limited relevance for today's practice. Lately, the results from the five-year follow-up of the larger randomized Syntax study have been published (7,8), which show better outcome in death and other hard endpoints in patients with complex coronary disease treated with CABG—thus, to a large extent, corroborating the results of the previous observational studies.

The present study comparing survival in two different time periods (1999–2005 and 2006–2011) revealed a small, but significantly higher survival rate in patients treated in the last time period with three-vessel disease, regardless of treatment strategy (HR: 0.80, 95% CI: 0.69–0.93, p = 0.004). It is difficult to pinpoint any single factor responsible for this increase in survival, but it may be related to a number of factors such as better equipment for both PCI and CABG, improvement in pre- and postoperative care and more optimal medical treatment.

The aim of the study was to assess whether the survival difference in patients with three-vessel disease treated with CABG compared with those treated with PCI noted in our first study (2) was different in the most recent time period. If so, it might indicate a need for reassessing the treatment strategy selection process. The most obvious change in PCI treatment between the two time periods was a substantial increase in the use of DES in era from 2006 to 2011 (12). The results, however, showed no indication of a diminished survival advantage in CABG-treated patients with three-vessel disease in the last time period. The analyses were quite consistent in revealing a clear reduction in HR for CABG-treated patients with three-vessel disease and no difference for one- and two-vessel diseases.

There are at least three study limitations that are noteworthy. First of all, it is an observational study with its inherent possibility of unmeasured selection bias. An attempt to minimize that influence was performed using different types of survival analyses: ordinary Cox regression, Cox regression using the propensity score as covariate, and propensity score matching. The results were consistent and in keeping with both previous observational and randomized studies. Secondly, the Syntax score (16) was not available in the database, which could have made a better definition of patients with complex coronary anatomy than the rather crude distinction between one-, two-, and three-vessel diseases. Thirdly, in view of multiple comparisons without any Bonferroni adjustment, the p values should be interpreted with caution.

In conclusion, the study reveals a consistent higher survival rate of CABG-treated patients compared with patients with three-vessel disease treated with PCI. These results are in agreement with most previous observational studies (1,3–5) and also with the recent results from the randomized SYNTAX study (7,8). There was no indication of a reduction in this benefit in the last time period. Thus, the standard treatment option for most patients with three-vessel disease should still be CABG.

Declaration of interest: The author reports no conflict of interest. The author alone is responsible for the content and writing of the paper.