Does coronary artery bypass surgery improve survival?

Abstract

Objectives. According to guide-lines, coronary bypass surgery improves survival in high risk patients. The evidence for this is more than 20 years old and may be questioned. Long waiting lists for coronary bypass surgery are detrimental but offer the possibility to compare the risk of death before and after surgery. We hypothesized that the risk of death is lower after bypass surgery than before the operation in high risk patients in a more recent cohort. Design and results. Death hazard functions were calculated by the use of Poisson regression scheduled for bypass surgery between 1 Jan 1995 and 31 July 2005. The analyses were performed in two states: 1) in the period after triage until admission for surgery during which optimal medication was intended and 2) after surgery and up to 11 years (corresponding to 57,548 patient years). The probability of death was calculated by entering individual risk profile data into the two multivariable functions. There were several significant differences between the hazard functions in the two states. All variables reflecting angiographic severity of coronary lesions indicated lower risk of death after bypass surgery. The risk associated with left ventricular impairment was lower after surgery (beta coefficients − 0.0546 vs. − 0.0234, p <0.001). Only one variable, age, indicated higher risk after surgery (which is also seen in a general population over time). The reduction of risk was dependent on preoperative risk with a large reduction when preoperative risk was high and vice versa. When preoperative risk was low, however, the risk increased due to surgical mortality. Conclusions. The risk of death is lower after bypass surgery than before the operation in high risk patients. This is most likely explained by a prognostic gain from bypass surgery. The gain is largest in high-risk patients but small or absent in low risk patients.

Key words::

• coronary
• bypass surgery
• mortality
• death hazard functions

Introduction

The major indications for coronary artery bypass grafting (CABG) are to relieve symptoms, to reduce the risk of myocardial infarction and to prevent death. This study focuses on the last indication. According to guidelines, for example, evidence based cardiology (1) ‘a strategy of early coronary bypass surgery improves long-term survival in a broad spectrum of patients at moderate to high risk compared with medical therapy. Absolute benefits are proportional to the risk expected with medical therapy.’ The evidence is to a large extent based on three moderately sized randomized trials (2–5) that were conducted more than 20 years ago. Pharmacological treatment was then different, the surgical technique was different, and the selection of patients differed from that of today. Therefore, the implications of these studies today are uncertain and should be re-assessed.

It could be argued that studies that have compared risk of death after CABG and after percutaneous coronary intervention (PCI) provide enough, albeit indirect, up-to-date evidence. The registry study by Hannan et al. (6) and the randomized SYNTAX study with 3-year follow-up data (7) clearly indicate that survival is better after CABG than after PCI (in SYNTAX cardiac deaths are less frequent). Isn't that enough to conclude that CABG still improves survival? Possibly, but from a strictly scientific point of view the observation that CABG is better than PCI does not prove that CABG reduces risk of death. How come? Because it has not been shown conclusively that PCI has any effect on survival compared to medical treatment (except perhaps in acute myocardial infarction). Therefore, one cannot exclude the possibility that PCI, as an invasive measure, instead increases the risk of death. If so, the difference between CABG and PCI could be explained by a negative effect of PCI and not by a positive effect of CABG.

Many would propose that new prospective randomized trials should be instigated to settle whether or not coronary bypass surgery today improves survival in comparison to pharmacological agents and life style changes alone. However, the randomized trial is less than ideal in the present context. First, there could be concern about the ethics of a randomized trial, where half of the patients are not offered a treatment that previously has been shown to improve survival. Second, a very large number of patients would have to be randomized, up to 8000 if moderate to high risk patients were included (8). It seems unlikely that such a large study will be pursued. But even if it would, the statistical power would be insufficient for detailed sub-group analyses. These are essential considering the large variation in effect shown in a meta-analysis of the early randomized trials (5,8). Due to these issues, conclusive prospective randomized studies are not likely to be undertaken.

In many countries over the years, the capacity for coronary bypass surgery has been insufficient to meet the demand (9–11). This has resulted in waiting lists for surgery with a varying delay of surgery of up to more than a year at our institution The regrettable situation, however, can be used to address the issue of survival benefit of coronary bypass surgery.

In the present study, 10,657 patients scheduled for isolated coronary artery bypass surgery at one institution and its affiliation between 1995 and July 2005 were included in an analysis that compared the death hazard functions in two states: 1) while on the waiting list for surgery and 2) after the operation. We hypothesized that the death hazard functions would differ with a lower risk of death after bypass surgery than before the operation in high risk patients.

Patients and methods

Patients

During the inclusion period (January 1995–31 July 2005), 11 261 patients were accepted for primary isolated CABG at Sahlgrenska University Hospital and Scandinavian Heart Center. The two centers have a joint waiting list, and the same surgeons operate at both facilities. Of these, 384 patients (3.4%) underwent an emergency surgical procedure (within 24 hours after acceptance), and therefore were never included on the waiting list and thus excluded from this study. The remaining 10 877 patients entered the waiting list. Of these patients, 220 (2.0%) were withdrawn from the waiting list and subsequently from the study for various reasons; such as patient declining surgery, change of treatment to PCI or concomitant serious illness. The final study population consisted of 10 657 patients accepted for elective, isolated, primary coronary bypass surgery. Preoperative data were collected at acceptance from the preoperative assessment by the treating cardiologist and were registered prospectively in a database (CorBase; Journalia AB, Kungälv, Sweden). Patient characteristics and amount of missing data are presented in Table I. Mortality data was collected from the Swedish National Tax Board (Skatteverket) and were complete. Deaths from all causes were reported. The patients were followed until death or to March 1, 2006. Mean follow-up time was 5.4±2.9 years (range 0–11 years) corresponding to 57,548 patient years.

Table I. Patient characteristics. Mean ± standard deviation or number (%).

		Missing data (n, %)
Number of patients (n)	10657	0 (0)
Men	8344 (78)	0 (0)
Age (years)	66±9	0 (0)
Waiting time (days, median and interquartile range)	39 (13–89)	0 (0)
Waiting time (days)	64±74	0 (0)
Body mass index	27±4	245 (2.3)
Cleveland clinic risk score	1.67±1.61	287 (2.7)
Left main stenosis	2261 (21)	189 (1.8)
Unstable angina	3177 (30)	92 (0.9)
Chronic obstructive pulmonary disease	563 (5)	380 (3.6)
Previous stroke	535 (5)	363 (3.4)
with previous TIA included	858 (8)
Hypertension	4498 (42)	363 (3.4)
Atrial fibrillation	215 (2)	370 (3.5)
Diabetes mellitus	2107 (20)	260 (2.4)
Serum creatinine (µmol/L)	102±49	293 (2.7)
Blood hemoglobin (g/L)	140±13	179 (1.7)
NYHA-class		370 (3.5)
I	6792 (64)
II	1943 (18)
III	1395 (13)
IV	157 (2)
CCS-class		488 (4.6)
0	977 (9)
I	588 (5)
II	2162 (20)
III	4893 (46)
IV	1549 (15)
Priority group		54 (0.5)
Imperative	3909 (37)
Urgent	4774 (45)
Routine	1919 (18)
Number of vessels diseased		189 (1.8)
Three-vessel disease	7429 (70)
Two-vessel disease	2354 (22)
One-vessel disease	684 (6)
Unknown	189 (2)
Ejection fraction	58.0±12.9	351 (3.3)

CCS, Canadian Cardiology Society; NYHA, New York Heart Association; TIA, transient ischemic attack.

Definitions

Waiting time was defined as the time from acceptance to operation, and for those who died while waiting, as the time from acceptance to death. However, in the statistical evaluation, waiting time at a certain moment was defined as the time since acceptance. Cardiac catheterization was performed within 1 week before triage in patients in stable condition and within the last 24 hours in patients with unstable angina. Significant stenosis was defined as a 50% reduction in vessel diameter as measured by angiography. Left ventricular ejection fraction (LVEF) was assessed by transthoracic echocardiography in the majority of patients and in the others with a left ventricular injection during coronary angiography. The severity of symptoms of cardiac failure was classified according to the New York Heart Association (NYHA class, Ref. 12).

Early mortality was defined as death from all causes within 30 days after surgery. Total postoperative mortality was all mortality until the end of the follow-up period (including 30-day mortality).

Triage

All patients were accepted and allocated to priority groups at triage by the treating cardiologist, a senior cardiothoracic surgeon and an interventional cardiologist. The decisions were mainly based on the severity of symptoms, extent of coronary disease, and left ventricular function. The elective patients were prioritized into three groups: A) imperative: surgery planned within 2 weeks, B) urgent: surgery planned within 12 weeks, and C) routine: the remaining patients. If priority was changed during the study period, the final priority was used in the analyses.

Statistics

The statistical aim of this study was to obtain death hazard functions based on two states: 1) while on the waiting list for surgery and 2) during the follow-up after surgery and to compare these functions. Therefore, the relationships between potential risk variables as known at triage (listed in Table II) and death events were analyzed by use of Poisson regression as previously described (11,13–15). Poisson regression was chosen for the analysis instead of the Cox proportional hazard model, which is the usual one applied when analyzing survival with several possible predictors. However, the proportional hazard requirement is that the quotient between the hazard functions before and after surgery should be the same independently of time since decision. This requirement is not fulfilled in our study as can be seen in Figure 1. In addition, the special type of Poisson regression results in continuous hazard functions over time allowing for a changing relation between function curves (not possible with Cox regression).

Table II. ‘Univariable’ analyses of hazard functions of death (‘univariable’ analysis of β–coefficients and hazard ratios is not truly univariable but includes time since TRIAGE and surgery respectively and age).

	Before surgery					After surgery
Variable (all variables are as known at triage)	β	SE	p–Value	HR	95% CI	β	SE	p–Value	HR	95% CI	p–Value (before vs. after surgery)
Gender	−0.0956	0.2778	0.7308	0.91	0.53–1.57	−0.0036	0.0628	0.9543	1.00	0.88–1.13	0.7467
Age	0.0334	0.0128	0.0091	1.03	1.01–1.06	0.0603	0.0033	0.0000	1.06	1.06–1.07	0.0419
Time on waiting list						0.3328	0.1154	0.0039	1.39	1.11–1.75
Body mass index	0.0135	0.0206	0.5123	1.01	0.97–1.06	−0.0033	0.0039	0.3975	1.00	0.99–1.00	0.4230
LVEF (left ventricular ejection fraction)	−0.0594	0.0075	0.0000	0.94	0.93–0.96	−0.0302	0.0019	0.0000	0.97	0.97–0.97	0.0002
Chronic obstructive pulmonary disease	0.7711	0.3370	0.0222	2.16	1.12–4.19	0.7045	0.0891	0.0000	2.02	1.70–2.41	0.8485
Preoperative stroke or TIA	0.8532	0.2851	0.0028	2.35	1.34–4.10	0.6678	0.0769	0.0000	1.95	1.68–2.27	0.5302
Hypertension	0.0900	0.2189	0.6810	1.09	0.71–1.68	0.3040	0.0523	0.0000	1.36	1.22–1.50	0.3417
Diabetes	0.9132	0.2224	0.0000	2.49	1.61–3.85	0.5542	0.0580	0.0000	1.74	1.55–1.95	0.1183
Smoker	2.2649	0.2466	0.0000	9.63	5.94–15.6	0.0571	0.0702	0.4160	1.06	0.92–1.21	0.0000
Preoperative heart failure	1.6694	0.2253	0.0000	5.31	3.41–8.26	1.0378	0.0667	0.0000	2.82	2.48–3.22	0.0072
Preoperative atrial fibrillation	1.3914	0.3717	0.0002	4.02	1.94–8.33	0.9432	0.1229	0.0000	2.57	2.02–3.27	0.2523
Preoperative unstable angina	1.6799	0.2285	0.0000	5.37	3.43–8.40	0.2934	0.0554	0.0000	1.34	1.20–1.49	0.0000
NYHA	0.5698	0.1303	0.0000	1.77	1.37–2.28	0.3315	0.0304	0.0000	1.39	1.31–1.48	0.0749
Se-creatinine	0.0025	0.0007	0.0004	1.00	1.00–1.00	0.0025	0.0003	0.0000	1.00	1.00–1.00	1.0000
Number of diseased vessels (1–3)	0.8188	0.2702	0.0025	2.27	1.34–3.85	0.3112	0.0501	0.0000	1.37	1.24–1.51	0.0647
Left main coronary disease	0.7868	0.2537	0.0019	2.20	1.34–3.61	0.1172	0.0615	0.0567	1.12	1.00–1.27	0.0103
Prox LAD-stenos	0.8777	0.2448	0.0003	2.41	1.49–3.89	0.1298	0.0519	0.0124	1.14	1.03–1.26	0.0028
3-Vessel disease	0.9212	0.3119	0.0032	2.51	1.36–4.63	0.3340	0.0612	0.0000	1.40	1.24–1.57	0.0647
3-Vessel disease + prox LAD	0.8928	0.2278	0.0001	2.44	1.56–3.82	0.2892	0.0516	0.0000	1.34	1.21–1.48	0.0098

Figure 1. To illustrate the difference between the pre- and postoperative risk models, we have chosen a moderately increased risk profile (age of 67 years at decision, EF =40%, diabetes, NYHA =2, S-creatinine =150 µmol/L, and stable angina operated 7 days after triage). The two curves give the preoperative (bold line) and postoperative (thin line) hazard function curves.

The analyses of death hazard functions were done in two steps and over the whole range of follow-up times. First, a ‘univariable’ analysis was performed to identify significant β-coefficients before and after surgery when considering time, age and a single risk variable. The β-coefficients from the pre- and post-operation states were tested for significant differences.

Variables with significant β-coefficients in the univariable analysis were tested in multivariable analysis. The hazard function h was of the form exp(β₀ + β₁·x₁ +… β_k·x_k), where the betas are coefficients and x₁, …, x_k are the values of the variables listed in Table 2 that have significant beta-coefficients. Some of the variables reflected time since decision or time since operation. The betas were estimated as well as the matrices of covariance of the betas (this procedure is similar to logistic regression and the Cox regression).

The statistical uncertainty of the beta coefficients is reflected by the matrix of covariance for each one of the models. The two matrices are based on the whole study population and comprise together 133 unique numbers. Therefore, when looking at a specific risk profile, the statistical power is based on the total study population, not on the number of patients with that particular risk profile (in contrast to in the randomized trial but in common with logistic and Cox regression).

The accuracy of the hazard functions was tested by comparing predicted and observed deaths during the whole range of follow-up.

To illustrate the hazard functions, the individual risk variable values were entered into both hazard functions to calculate the potential probability of death before and after the operation for each individual patient (probability of death =1-exp(− the area under the hazard function curve for the chosen time period) ≈ the area)). The time of follow-up was set to 4 months in both the preoperative and the postoperative hazard function (in spite of a considerably longer actual follow-up in both instances). The time was restricted to 4 months since the calculations were then based on a large number of patients in both the preoperative and the postoperative state to ensure the validity of these calculations. The results are presented as means (of probability of death) for patients of various risk categories.

In the final preoperative multivariable risk model, 10,031 patients are included and in the final postoperative multivariable model 7209 patients are included.

Ethical aspects

The study complies with the declaration of Helsinki and was approved by the Research Ethics Committee at the Medical Faculty, University of Göteborg.

Results

Mortality

During the study period, 85 patients died while waiting for coronary bypass surgery (0.8%), corresponding to a case fatality rate of 46 deaths/1000 patient years on the waiting list. A total of 170 patients died within 30 days after surgery, resulting in an early mortality of 1.6%. Total postoperative mortality during the follow-up period (mean 5.4 years) was 14.7% (n =1552).

Death hazard functions

The variables with significant β-coefficients before and after surgery are given in Tables II (univariable analysis) and 3 (multivariable analysis). There were several significant differences between the hazard functions before and after surgery. All variables reflecting angiographic severity of coronary lesions indicated lower risk of death after surgery (3/5 p <0.02 and 5/5 p <0.07). The risk associated with left ventricular impairment (p =0.0002), heart failure (p =0.007), and NYHA class (p =0.07) were lower after surgery. Only one variable, age, indicated higher risk after surgery (which is also seen in a general population over time).

In multivariable testing before surgery, proximal LAD stenosis, left ventricular ejection fraction (LVEF), unstable angina, serum-creatinine, and diabetes were significant while after surgery age, preoperative heart failure, atrial fibrillation, NYHA class, LVEF, diabetes, chronic obstructive pulmonary disease (COPD), stroke, or transient ischemic attack (TIA) and serum-creatinine were significant (Table III). The importance of ejection fraction (in the multivariable testing) was more pronounced before surgery compared to after surgery (beta coefficients –0.0546 vs. –0.0234, p <0.001).

Table III. Multivariable analyses (significant β–coefficients, age, and time since TRIAGE and surgery respectively were included).

Variable (as known at triage)	β	SE	p–Value	HR	95% CI
Before surgery
Constant	−3.0822	1.0482
Current age	0.0202	0.0135	0.1358	1.02	0.99–1.05
EF	−0.0548	0.0079	0.0000	0.95	0.93–0.96
Diabetes	0.8762	0.2345	0.0002	2.40	1.52–3.80
Unstable angina	1.3517	0.2544	0.0000	3.86	2.35–6.36
S–creatinine	0.0035	0.0008	0.0000	1.00	1.00–1.01
Prox LAD–stenos	0.7534	0.2622	0.0041	2.12	1.27–3.55
Time since decision	−0.2443	0.5762	0.6716	0.78	0.25–2.42
After surgery
Constant	−4.2938	0.3572
Min(time, 0.1)	−25.3185	2.2868	0.0000	0.00	0.00–0.00
Max(Min(time-0.1, 0.5–0.1), 0)	−1.9103	0.5793	0.0010	0.15	0.05–0.46
Max(Min(time-0.5, 2–0.5), 0)	0.1451	0.1025	0.1569	1.16	0.95–1.41
Max(time–2, 0)	0.1210	0.0175	0.0000	1.13	1.09–1.17
Current age	0.0550	0.0042	0.0000	1.06	1.05–1.07
Left ventricular ejection fraction (LVEF)	−0.0234	0.0025	0.0000	0.98	0.97–0.98
Chronic obstructive pulmonary disease	0.5623	0.1109	0.0000	1.75	1.41–2.18
Preoperative stroke or TIA	0.4488	0.0961	0.0000	1.57	1.30–1.89
Diabetes	0.4748	0.0725	0.0000	1.61	1.39–1.85
Preoperative heart failure	0.3645	0.0901	0.0001	1.44	1.21–1.72
Preoperative atrial fibrillation	0.4069	0.1360	0.0028	1.50	1.15–1.96
NYHA	0.1315	0.0377	0.0005	1.14	1.06–1.23
S–creatinine	0.0029	0.0003	0.0000	1.00	1.00–1.00

Variables reflecting time since TRIAGE and time for bypass grafting respectively were included in the analysis. In the state before the operation, time had no significant impact, that is, no change in risk, up or down, was detected during this period. In the postoperative state three phases could be discerned. In the early postoperative period risk increased. Then during the subsequent half year it decreased, and in the longer perspective it increased again at a magnitude that is expected from increasing age.

The accuracy of the multivariable hazard functions was evaluated by comparing expected and observed mortality during the entire follow-up of the respective hazard-functions (Figure 2). As the figures make clear, the deviation between expected and observed mortality was marginal. This implies that the hazard functions are valid. However, for the preoperative hazard function this statement is limited to the first half-year of follow-up.

Figure 2. Predicted and observed deaths (absolute numbers) before (to the left) and after surgery (to the right) accumulated over time.

Probability of death

To illustrate the hazard functions and the difference between preoperative and postoperative functions, the individual risk variable values of all patients were entered into the respective hazard function. The time of follow-up was set to 4 months to ensure the validity of calculations.

The probability of death was lower after surgery than before in all high risk categories. In contrast, in low risk categories the risk increased due to surgical mortality. The risk varied between 32% and 0.2% before surgery and between 6.6% and 0.8% after surgery. The largest differences were seen when the preoperative risk was high and vice versa. The results of all patients are given as means of various risk categories in Table IV.

Table IV. Probability of death (Individual values of all patients were entered in the two hazard function and the probability of death for each patient is presented according to the criteria of table headings).

						Probability of death (%) in the interval within 4 months
Ejection fraction	Diabetes	Unstable angina	Prox LAD stenosis	Number of patients	Proportion of patients (%)	Not operated	Operated
EF ≤30	No	No	No	49	0.7	1.8	2.8
EF ≤30	No	No	Yes	75	1.0	3.9	3.0
EF ≤30	No	Yes	No	27	0.4	6.4	3.4
EF ≤30	No	Yes	Yes	66	0.9	13.9	4.0
EF ≤30	Yes	No	No	28	0.4	3.9	4.3
EF ≤30	Yes	No	Yes	24	0.3	8.4	4.9
EF ≤30	Yes	Yes	No	23	0.3	16.1	6.6
EF ≤30	Yes	Yes	Yes	21	0.3	32.2	6.1
30 < EF ≤50	No	No	No	416	5.8	0.7	1.6
30 < EF ≤50	No	No	Yes	483	6.7	1.5	1.7
30 < EF ≤50	No	Yes	No	203	2.8	2.7	1.9
30 < EF ≤50	No	Yes	Yes	305	4.2	5.8	2.0
30 < EF ≤50	Yes	No	No	110	1.5	1.9	3.1
30 < EF ≤50	Yes	No	Yes	168	2.3	3.7	2.8
30 < EF ≤50	Yes	Yes	No	74	1.0	6.1	3.5
30 < EF ≤50	Yes	Yes	Yes	97	1.3	13.4	3.3
EF >50	No	No	No	1413	19.5	0.2	0.8
EF >50	No	No	Yes	1619	22.4	0.5	0.9
EF >50	No	Yes	No	487	6.7	1.0	0.9
EF >50	No	Yes	Yes	620	8.6	2.0	1.1
EF >50	Yes	No	No	324	4.5	0.6	1.5
EF >50	Yes	No	Yes	353	4.9	1.2	1.5
EF >50	Yes	Yes	No	124	1.7	2.2	1.5
EF >50	Yes	Yes	Yes	122	1.7	4.9	1.8

Notice that more variables than ejection fraction, diabetes, unstable angina, and prox LAD (used in Figure 1 and Table IV) are included in the model. That means that within a group the other variables such as age, S-creatinine, etc., vary.

Discussion

Our results are in line with the early randomized studies (3–5), which indicated that coronary bypass surgery can reduce the risk of death in patients with coronary artery disease. More than anything, it illustrates the large variability in benefit between patients. Thus, patients with a high preoperative risk had a greater benefit from surgery than those at lower risk.

Admittedly, analysis of a patient series gives less reliable data than adequately sized prospective studies. But in the absence of such a study of recent origin, the present results are in close harmony with previous knowledge obtained decades ago and could be useful and supportive in clinical practice.

Death hazard functions

The beta-coefficients, which estimate the association between a variable and risk of death, were mostly lower after surgery. Of particular interest are the beta-coefficients that directly reflect coronary stenoses. One would expect that effective bypassing would alter the association between these variables and the risk of death. Hence, the observation that the corresponding β-coefficients were lower after surgery is clearly in line with our hypothesis.

Ventricular impairment is also of interest since it is associated with increased risk of death and also with the effects of bypass surgery. The early randomized studies found that risk reduction was greater in patients with preoperatively impaired function. This study includes three preoperative variables that reflect ventricular impairment directly (LVEF) or indirectly (preoperative period with heart failure and preoperative NYHA-class). The beta-coefficients of all three decreased after surgery (two of them with p <0.01 and one with p =0.07). The values known at triage were used. We do not know if the values were different after surgery. One may speculate that the reduced impact of heart function variables may reflect postoperatively improved function. If not, the observation indicates that bypassing reduces the risk of death in patients with preoperatively impaired heart function.

Smoking is a major risk factor for cardiovascular disease. The beta-coefficient for smoking decreased markedly and was not significant after surgery. Again, the information known at triage was used. Many patients stop smoking after bypass surgery, up to 50% has been reported recently (16). Therefore, it is possible that the lower β-coefficient is related to a lower number of smokers after surgery. An alternative explanation is that our finding is an example of the ‘smokers’ paradox’ (17). At any rate, preoperative smokers seem to gain more than preoperative nonsmokers, other risk variables being the same.

The only beta-coefficient that increased after surgery was age (and time during the early postoperative period, which was later reversed into a decrease of risk). In a general population, increasing age is naturally associated with a higher risk of death. This association is disrupted in severe, life threatening disease. It is possible that the increase of this variable may reflect a change towards normality, which would be in line with our hypothesis.

Probability of death

These considerations suggest that the probability of death could be lower after surgery. But actual risk during a time period is determined, not only by constants but also by individual variables, that is, the individual risk profile of the patients and the time of follow-up. The time is important, since most invasive procedures, such as coronary bypass surgery, carry a risk of death in conjunction to the procedure. Thus, an intervention which may aim at reducing risk of death is instead followed by an increased risk in a short perspective (Figure 1). After the postsurgical early period with elevated risk, the hazard function curves may intersect (which happened in less than 4 months after surgery in 67% of the patients in this study), so that the postsurgical function curve ends below the one before surgery. As illustrated in Figure 1, the area under the risk curve is larger early after surgery but after the curves have intersected it will be a matter of time before the preoperative area is larger than the preoperative.

For the comparison of probability of death between the two states we chose 4 months for the time of follow-up, since this period comprised a large number of patients in both the preoperative and postoperative status to enhance the accuracy of prediction. The very small deviation, in this time span, between expected deaths (as calculated by the hazard functions) and observed deaths (Figure 2) indicates a very good match, and supports the validity of the functions.

The downside of setting the time of follow-up to such a short time as 4 months is that surgical mortality may overshadow any beneficial effect of CABG. In the meta-analysis of early studies, it was not until 2 or 3 years that a positive effect on survival could be demonstrated, and the effect peaked between 5 and 7 years. As surgical mortality gets a disproportionate impact in our time frame, it is likely that we have underestimated potential benefit in a longer perspective. Even so, our data indicate that patients of high risk categories gained from the surgical procedure in terms of estimated reduction of risk. It seems possible that also patients at moderate risk will gain in a longer time perspective.

Study design and limitations

The study concept was to analyze risk of death functions in a patient series at two different states. The first state, the period between triage and surgery is a state where the intention was to achieve optimal medical treatment. It could be argued that this intention may not have been fulfilled, since the alternative invasive treatment had already been decided upon. However, the patients, with few exceptions, suffered from disabling pectoral angina and risked fatal myocardial infarction. The cardiologists were in charge of the treatment until the end of this period – when patients were admitted to the surgical ward for surgery. It is difficult to conceive that they, well aware that the patients would have to wait for weeks, months, or even more than a year until surgical treatment could be pursued, would consider a less than optimal treatment. So it is probable that the patients were on optimal treatment during this state.

The second state starts at surgery and continues until death or to the end of follow-up (specified in the ‘Patients’ section). The intention was still to pursue optimal medical treatment.

Our analysis showed that the risk functions of the two states were different, with a generally higher risk of death in the state before surgery. Had there been no other difference between the two states than the surgical procedure, the difference could have been attributed to surgery. However, there were other differences. First, time and age obviously differed, since the preoperative state preceded the postoperative. Second, the analyses are based on a different number of patients. And third, there may have been unknown factors that may explain the difference in risk functions. The alternative explanations to the lower risk after surgery have to be ruled out, before conclusions are made.

The possibility that the lower risk after surgery could be the result of time or age has to be considered. Would it have happened also without revascularization? In patients with stable angina (70% of patients), who were presented for surgery after optimal medical treatment, it is not likely that the risk decreased markedly within a few months without further intervention. Therefore, it is probable that the lower risk instable angina was caused by coronary bypass surgery. It may be different in patients with unstable angina. They either develop myocardial infarction, possibly fatal or stabilize at a lower risk level. So for this category our calculations may have overestimated the reduction in risk caused by coronary bypass surgery.

Furthermore, the analyses included time since triage and age as possible predicting variables. According to the analyses neither had any effect on risk of death while waiting for surgery. In all these considerations strongly suggest that time or age cannot explain but marginal differences.

The two hazard functions are based on different numbers of patients. This was caused, largely, by more missing data in the postoperative multivariable analysis, which were due to a greater number of significant risk variables. Therefore, the analysis of the preoperative state included more than 10,000 patients while there were 2828 less in the postoperative analysis. Did the loss of almost 3000 patients change the proportion of patients with different risk profiles and, if so, does this explain the difference in hazard function? This is not likely. The statement is based on two considerations. First, we have analyzed the hazard ratios of death of the almost 3000 patients that were not included in the postoperative analysis against the 7209 that were included. When adjusting for age only, the ratio was 1.0005 (95% CI: 0.90–1.11), that is, virtually identical risk. When adjusted for all variables of the preoperative hazard function the ratio was 0.984 (95% CI: 0.877–1.104, p =0.7870), that is, no difference is indicated. This implies that the loss of patients in the postoperative analysis had no marginal effect.

A second consideration is important. The primary aim of our analysis has been to obtain death hazard functions, that is, the relationship between risk variables and death. The specific patient mix is not critical for a valid analysis. What matters is that the postoperative analysis included the full range of values of the preoperative analysis and that all segments within the range were represented adequately. Our analysis has shown that these preconditions are met (data given at request). So, even if the nearly 3000 patients who were excluded in the postoperative analysis had differed (which they did not) this would not have changed the results.

In contrast, it is important when comparing probabilities of death (e.g., in terms of proportions of patients with different risk profiles that have much to gain and vice versa) that the patient mix is equal. In these calculations only the 7209 with a complete set of variables were included (naturally since this is a prerequisite for the calculation). Therefore, the patient mix was identical before and after operation.

Is it possible, that other, to us unknown factors/variables were important to explain differences in risk? It certainly is possible, but judging from the close match between observed deaths in the two states and the predicted risks, their importance was probably none or marginal. The match could hardly be much improved.

Finally, it should be noticed that the prioritization process and subsequent differences in waiting times may cause bias, since waiting time is not randomly allocated among the patient on the waiting list. Left main stem coronary artery lesion, for example, was not significant in the preoperative multivariable analysis, whereas proximal LAD stenosis was. This may seem curious but is most likely explained by the fact that left main stem lesions were given highest priority at triage (unlike proximal LAD stenosis) and the patients were almost without exception operated within a week or two. The very short waiting times make the analysis underestimate the risk of left main lesions.

Conclusions and implications

The present study of 10,657 patients scheduled for coronary bypass surgery shows that death hazard functions differ with a lower risk of death after bypass surgery than before the operation in high risk patients. The findings are compatible with the hypothesis, that the operation provides survival benefit for patients with high risk profile. Our data reinforce current guideline recommendations that advocate initial surgery in high-risk patients to save lives as well as for angina relief. In contrast, in patients with low or moderate risk preoperatively, the gain was small, absent, or even negative. Hence, the main indication for these patients is to relieve angina. With this objective our data support guidelines which advocate an initial approach in low risk patients with intensive medical therapy, a reduction of risk factors, and life style intervention in low risk patients. If these measures fail, bypass surgery should be considered.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper. The study was supported by Sahlgrenska University Hospital, (LUA/ALF-grant). The study sponsor had no influence on study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the report for publication.