Author + information
- Received March 24, 1998
- Revision received August 28, 1998
- Accepted October 2, 1998
- Published online February 1, 1999.
- William O Myers, MD, FACCa,* (, )
- Eugene H Blackstone, MD, FACC∗,
- Kathryn Davis, PhD, FACC†,
- Eric D Foster, MD, FACC‡ and
- George C Kaiser, MD, FACC§
- ↵*Reprint requests and correspondence: Dr. William O. Myers, Marshfield Clinic, 1000 North Oak Avenue, Marshfield, WI 54449
To show the effect of clinical, angio and demographic traits on late survival of Coronary Artery Surgery Study (CASS) patients following coronary artery bypass grafting (CABG) and introduce Hazard Function analysis to CASS survival data.
Patients were reached by mail survey with 94% response. By National Death Index, vital status was obtained in 99.7% (n = 8221) with a mean follow up of 15 years. Cox proportional hazard and Blackstone Hazard Function regressions were used to assess effects of preoperative traits.
Ninety percent of patients were alive at 5, 74% at 10 and 56% at 15 years. Of those age 65 and age 75 at operation, 74% and 59% were living at 10 years and 54% and 33% at 15 years (now age 90), survival exceeding the matched U.S. population. Hazard Function falls rapidly after CABG to 9 to 12 months, then rises, doubling by 15 years. Young patients, below age 35, had lower late survival. The time-segmented Cox model (divided at time suggested by the Hazard Function) identified traits showing predictive power early, throughout and late. Female sex, small body surface, ischemic symptoms and emergency status affected survival early. Heavier weight, infarct(s), diuretics, diabetes, smoking, left main and LAD stenosis and use of vein grafts only increased hazard late only.
There are still lessons from the CASS database. CABG in the elderly is supported by the survival pattern of our patients age 75 at operation. Time-segmented Cox analysis and Hazard Function analysis separate baseline variables into those that predict early mortality and those that predict long survival.
Our presentation of the Coronary Artery Surgery Study (CASS) surgical survival in 1985 (1)was intended as a natural history report without any comparison of survival with medical treatment. The randomized portion of the CASS and certain later registry observational studies served the medical-surgical comparison purpose. The present report is also a natural history study. We noted in 1985 that more than a million patients had been operated on in the U.S. by that time and a stand-alone natural history report for coronary artery bypass grafting (CABG) patients was needed. In the interval between that study and the present review, millions more have undergone CABG; a quarter of a million cases from 1997 were added in the Society of Thoracic Surgeons Database annual harvest alone (according to the Forward by Frederick L. Grover, MD, Data Analyses of the STS National Cardiac Surgery Database, January 1998).
The CASS data base is two decades old, and surgical techniques of that era were somewhat, but not a lot, different from today. The major advance is the use of arterial grafts, chiefly the left internal thoracic artery anastomosed to the left anterior descending coronary artery (LIMA-LAD). Follow up for that operation is addressed in this report, but there are still many surviving patients from the CASS era without arterial grafts for whom the present analysis is worthwhile information, as well as for the LIMA-LAD patients. The CASS data base is very large; the patients’ entering traits were thoroughly characterized, the follow up was very complete, and there are still lessons to be learned from it.
Patients and methods
Study group and followup
The CASS Registry is a database of 24,959 well-characterized patients entered between 1974 and 1979 from 15 centers across North America. Patients were followed annually through 1982 and then by a final mail survey between 1988 and 1991. Overall, 94% responded. Vital status through 1989 among nonresponders at last follow-up was obtained from the National Death Index and, in some cases, from next-of-kin, such that the status of 95.8% of all CASS patients was known.
Among the Registry patients, 8,248 underwent a first time CABG operation during follow-up through 1982 (so-called primary isolated CABG). Vital status was obtained in 8,221 (99.7%). The common closing date for follow up among survivors was December 31, 1992.
The mean follow-up interval among survivors was 15.4 ± 1.39 years. Among them, 90% were followed 13.7 years or more, and 10% were followed 17 or more years.
Characterization of overall mortality
Nonparametric estimates of overall nonrisk-adjusted survival were obtained by the method of Kaplan and Meier (2). A completely parametric method was used to resolve the number of hazard phases, identify the form of the equation for each phase and estimate the parameters that characterized the distribution of times until death (3).
For orientation an age-sex matched U.S. population life table was constructed by generating a predicted general population survival curve for each patient (conditional on being alive at time zero, the age at surgery) and finding the mean across time either overall or, in the case of stratification, by stratum.
Multivariable analyses of death
Values for variables were obtained from studies closest to the date of surgery, as long as they were obtained not more than three years before surgery. The potential risk factors (variables) were organized for entry into the analysis as in Appendix A.
General conduct of the multivariable analyses
Exploratory analysis included correlation analysis, univariable association of risk to each variable and decile risk analysis of ordinal and continuous variables to identify possible transformations of scale. A directed technique of stepwise entry of variables into the multivariable risk factor model was then used (4). The p-value criterion for retention of variables in the final analysis was 0.05. Patients with missing data were removed from the analysis except when specifically noted. Regression coefficients are presented plus or minus one standard error.
A completely parametric multivariable analysis of death was made using hazard function methodology (available from ftp://uabcvsr.cvsr.uab.eduvia the Internet) (2). In this analysis, the variables were entered simultaneously into the scaling parameter of each phase of the mathematical equation characterizing the underlying distribution of times until death.
Five semiparametric Cox proportional hazards regression analyses were conducted (5). The first was an overall analysis, and the others were time-segmented analyses. For the latter, the hazard function was used as a guide to determine approximate time points for the end of the early phase of hazard and the beginning of the late phase (see Appendix B). This occurred at approximately 9 to 12 months. Therefore, piece-wise time-segmented Cox analyses were performed for two time periods. For one pair of analyses, deaths occurring within the first 9 postoperative months were analyzed, with follow up beyond that time set to 9 months; then deaths beyond 9 months were analyzed, with patients with follow-up intervals (and events) within the first 9 months deleted. A second pair of analyses along the same lines was performed but with the “cut-point” at 12 months.
Nature and influence of risk factors
Exploration of the influence of risk factors in the parametric multivariable analysis was performed by constructing a series of nomograms representing the solution of the parametric equation for specific supplied values of each factor. Thus, each figure represents the prediction for a specific, but typical, hypothetical patient. Each plot was supplemented by a population life table matched to that hypothetical patient.
Among the 8,221 patients whose vital status was determined, 3,769 died. Overall survival was 98%, 96%, 90%, 74%, 56% and 45% at 30 days, 1, 5, 10, 15 and 18 years respectively (Fig. 1). The instantaneous risk of death (the hazard function) was found to have two hazard phases. The first was a declining hazard phase from the time of operation (early hazard) throughout nearly the first postoperative year (see Appendix B); it accounted for 332 deaths. It then gave way to a steadily rising hazard phase (late hazard) that continued to rise throughout the reminder of follow-up and accounted for 3,437 deaths. The magnitude of hazard was higher than that for the age-sex matched general population.
Multivariable analyses of death
Patient and procedure risk factors for death at any time during follow up in the multiphase hazard function domain are listed in Table 1. Several factors appearing in both hazard phases have approximately the same regression coefficient magnitude in each phase, indicating that they influence survival proportionally across follow up time (left ventricular [LV] wall motion score, presence of hypertension and use of digitalis at baseline). Other factors appearing in both hazard phases have a variable, nonproportional influence across time (age, number of diseased systems and LV end diastolic pressure). Yet other factors appear only in the early or late hazard phase, indicating a nonproportional influence of those variables across time.
The parametric analysis yielded a solvable mathematical equation that can be used to generate time-related survival curves for individual patients.
The overall Cox analysis (not shown) identified the majority of the variables identified by the parametric hazard regression method, including the same transformations of ordinal and continuous variables. For these variables, the regression coefficients were of similar magnitude and degree of certainty. However, four of the early hazard phase variables and one of the late hazard phase variables were different. In the cases of the Canadian Heart Class and surgical priority, the Cox analysis identified a different variable and transformation from that selected from the same category of information in the hazard function regression analysis. Body surface area, inferior wall jeopardy and being a previous smoker appeared only in the hazard function analysis.
In contrast, the two pairs of time-segmented proportional hazards Cox models identified the same risk factors found in the multiphase hazard analysis, except that in the analysis with a cut-point of 9 months, body surface area is not identified (p = 0.07). The absence of surface area results has the effect of increasing the influence of female sex. (Table 2shows the model with cut-point at 9 months.)
The regression coefficients were of similar magnitude in the parametric and segmented semiparametric types of analyses, particularly in the late hazard phase, and both types differed from the overall Cox analysis in the same fashion.
Risk factors for death after operation
Both older, and even more so, younger, age at operation were profoundly associated with lower survival (Fig. 2). The younger the patient was below about age 35 years, the more severely reduced was survival. Likewise, patients older than about 70 years had a lower survival; however, this appeared to be better than that expected for the similarly aged general population.
Larger, heavier patients were at somewhat increased risk, particularly late. In contrast, early after operation smaller, lighter patients were at slightly greater risk.
Women were at higher risk than men and this risk was not accounted for by adjusting for differences in prevalence of other risk factors (Fig. 3). However, in the general population, women are at less risk than men, so that the discrepency between survival after CABG and that for the general population is magnified in women. In general, women coming to operation are more symptomatic, though they have lesser disease and better preserved left ventricular function (6).
Symptoms of reversible ischemia
The more severe the symptoms of reversible ischemia at operation (anginal class), the higher was the early risk. However, the difference is small in the typical patient, in part because the overall risk in the early hazard phase is small, and the overall impact of the increment in risk from more severe symptoms is negligible.
Left ventricular (LV) function
The more severe the depression in LV wall motion, the higher the early and late risk. In this study global ejection fraction was not available for a large number of patients, and this missing information was, for some institutions, systematic for large heart sizes. The use of wall motion abnormality alone was found satisfactory when ejection fraction is above about 35%. However, when the ejection fraction is severely depressed, survival was somewhat worse than that predicted by wall motion score.
The additional increment of a previous myocardial infarction to that already accounted for by wall motion abnormalities is small. Severely elevated, but not modestly elevated, LV end diastolic pressure seen in a relatively small number of patients had progressively adverse influence on survival as time passed. Patients on digitalis for any reason (Fig. 4)and, to a lesser extent those using diuretics, were at greater risk in the late follow up period.
Distribution and extent of coronary artery disease
Despite the operation, the greater the degree preoperatively of coronary artery disease, the less favorable the survival, early and late. By risk-unadjusted life table analysis, the difference in survival according to number of diseased systems seems to be large and the difference in risk-adjusted survival appears more modest. The apparent larger differences are due to a higher prevalence and severity of other risk factors among patients with more extensive coronary disease (Table 3).
The location of disease was also a risk factor, with anterior disease placing the patient at somewhat greater risk early on. However, the long-range effect of this risk factor is negligible. The influence of left main or proximal LAD disease though unlikely to be due to chance, was small, indicating that the operation had by-andlarge neutralized these important natural history risk factors (Table 4).
Coexisting conditions influenced predominantly late events, and their effects were often considerable. Coexisting diabetes (Fig. 5), and to only a somewhat lesser degree, hypertension (Fig. 6), was associated with considerably increased risk for late death. If the patient was a smoker at the time of CABG, the late risk was higher than if the patient had never smoked or had been a previous smoker (Fig. 7). There was a trivial decrease in survival among previous smokers.
Surgical priority in this earlier era of coronary artery bypass grafting probably reflected the degree of perceived or actual clinical precariousness of the patient. It is reflected in a several fold increased early mortality, particularly when the operation was done under emergency conditions, although the overall impact on survival was not great.
The internal thoracic (mammary) artery (IMA) was used as a bypassing conduit (generally to the LAD system) in only 13% of patients, in contrast to the current era. The patient profile of these patients did not differ importantly from those undergoing operations in which only vein grafts were used as bypassing conduits. Early surgical survival was not adversely affected by the placement of such a conduit. Late survival was better among patients who received such a conduit than among those receiving vein grafts only (Fig. 8).
Comparison with previous surgical registry survival paper
This study extends our analysis of survival of surgical patients to beyond fifteen years. The previous study in 1985 (1)followed 8,981 patients who had coronary bypass surgery from 1974, when CASS began enrollment up until 30 days before constructing the file for the statistical analysis for that paper, a period of eight years, providing a five year follow up for the great majority of patients in that study. The present paper, 10 years later, follows those patients out beyond 15 years. Randomized and randomizable surgical patients (7)are included. Some patients (we are short 733 or 8% of those analyzed before) were removed because of the requirement for this study for the baseline data to have been collected within three years (prior) to the CABG operation. (The previous study allowed older baseline data.) Otherwise the inclusion and exclusion criteria are essentially the same: first time CABG only, no concomitant operation whether cardiac or not (noncardiac concomitant procedures were not specifically excluded before), no congenital heart lesions (an operation for a congenital lesion was the excluder in the older study).
Observational study problems
Some criticisms can be directed to this paper: first, the era of surgical practice represented. Only 13% of surgical patients had arterial grafts (internal thoracic artery or IMA grafts) but they constituted a large enough group to contrast long term survival to those without IMA and to compare with the age and sex matched U.S. population. Thousands of CABG late survivors from the CASS era have no arterial grafts, but the effects of other variables we report here should be valuable information in their clinical management. The nomogram of Figure 2and others that follow (with no IMA) supports the benefits of vein as conduit, although greater advantage was conferred by use of the IMA.
Patients over age 65 were not randomized to treatment in CASS and are usually excluded from randomized trials. The U.S. population, age and sex matched, is included for a reference point; it is interesting, and we don’t have anything better. We recognize the limitations of such a comparison.
Other CASS late surgical papers
This paper takes its place with several other CASS reports of this decade to round out the story of the CASS: (6–14), including the 10-year survival and myocardial infarction follow up of the randomized trial by Alderman et al. (8), the quality of life 10-year report also of the randomized group by Rogers et al. (9), two other randomized reports by Chaitman et al. (10)and Weiner et al. (11), a late medical follow-up paper by Emond et al. (12)and a late follow up of women in CASS by Davis et al. (13). A 15 year follow up of CABG with internal thoracic artery grafts was reported by Cameron et al. in 1996 (14), updating a previous report from 1988.
This report introduces hazard function analysis to the CASS long term follow up, providing further insight into which patient traits are continuous hazards, which operate only early and which operate mainly late.
In our earlier study, we compared surgical survival in other published series that we found similar to the CASS Registry. Some have since reported longer follow up. For example, the Duke University group reported a 20-year study concluding that higher risk patients with more severe disease should be considered for CABG because surgery had the greatest affect on reducing cardiovascular events in such patients (15). The Baylor group reported patients followed 15 to 20 years with 20% of their three-vessel disease patients surviving at 20 years (16). The Portland group, also reporting 15 to 20 year data, showed a 38% survival overall with a 33% actuarial reoperation rate at 15 years (17). We make no attempt to evaluate reoperation rates in the present study because CASS had a several year hiatus in detailed follow up, and we have no idea of patient symptoms or treatment in the interval before the vital status assessment of this study.
Although comparison with the U.S. age- and sex-matched population has well known problems, we believe coronary bypass for elderly ischemic heart disease patients who are otherwise suitable surgical candidates is supported by this study.
Variables considered in multivariable analysis of death
Gender; age (years) at operation; weight; height; body surface area; body mass index; weight-height ratio, obesity index
Symptoms of reversible ischemia
Presence of angina, Canadian Heart Class (Angina Class augmented to 5 for nonexertional angina), Canadian Angina Class, angina unrelated to exertion, unstable angina, augmented Canadian Angina Class (unstable angina as Class 5), chronicity of symptoms, use of nitroglycerin.
Left ventricular function
History of previous myocardial infarction(s), history of one previous infarct, history of two or more infarcts, history of congestive heart failure, presence of rales on physical exam, presence of S3 sound, use of digitalis, use of a diuretic, CASS left ventricular wall motion score, left ventricular end diastolic pressure.
Distribution and extent of coronary artery disease
Coronary system dominance, number of coronary systems diseased (70% diameter reduction or more), number of proximal coronary arteries diseased, left main disease (%), proximal LAD disease (%).
Diabetes (treated), hypertension (treated), smoking history (ever smoked, previous smoker, current smoker).
Elective, urgent, or emergent operation.
Use of saphenous vein conduits only, use of internal thoracic (mammary) artery graft(s) with or without other vein grafts.
Hazard function components
Two hazard phases were resolved, an early declining and a late rising phase. These sum to the overall hazard function (Figure 9). The relative influence of each hazard phase across time is represented by the percent of total hazard at each point in time contributed by the early and by the late hazard phases.
☆ This study was supported by Grant RO1-HL38941 from NIH, Bethesda, Maryland.
- Coronary Artery Surgery Study
- coronary artery bypass grafting
- left anterior descending
- internal mammary artery
- left ventricular
- left internal mammary artery to left anterior descending
- Received March 24, 1998.
- Revision received August 28, 1998.
- Accepted October 2, 1998.
- American College of Cardiology
- Cox D.R
- Alderman E.L,
- Bourassa M.G,
- Cohen L.S,
- et al.
- Rogers W.J,
- Coggin J,
- Gersh B.J,
- et al.
- Weiner D.A,
- Ryan T.H,
- Parsons L,
- et al.
- Emond M,
- Mock M.B,
- Davis K,
- et al.
- Davis K,
- Chaitman B,
- Ryan T,
- Bittner V,
- Kennedy J.W
- Rahimtoola S.H,
- Fessler C.L,
- Grunkemeier G.L,
- Starr A