Journal of the American College of Cardiology
Prognostic factors for atherosclerosis progression in saphenous vein graftsThe postcoronary artery bypass graft (post-CABG) trial
Author + information
- Received February 28, 2000
- Revision received June 22, 2000
- Accepted July 31, 2000
- Published online November 15, 2000.
Author Information
- Michael J. Domanski, MD, FACC∗,*,
- Craig B. Borkowf, PhD†,
- Lucien Campeau, MD, FACC‡,
- Genell L. Knatterud, PhD§,
- Carl White, MD, FACC∥,
- Byron Hoogwerf, MD¶,
- Yves Rosenberg, MD, MPH∗,
- Nancy L. Geller, PhD†,
- the Post-CABG Trial Investigators 8#
- ↵*Reprint requests and correspondence: Dr. Michael Domanski, National Heart, Lung, and Blood Institute, Bethesda, Maryland 20892
Abstract
OBJECTIVES
The study was done to assess patients in the Post-Coronary Artery Bypass Graft (Post-CABG) trial to determine prognostic factors for atherosclerosis progression.
BACKGROUND
Saphenous vein grafts (SVGs) are effective in relieving angina and, in certain patient subsets, in prolonging life. However, the progression of atherosclerosis in many of these grafts limits their usefulness.
METHODS
The Post-CABG trial studied moderate versus aggressive lipid-lowering and low-dose warfarin versus placebo in patients with a history of coronary artery bypass surgery and found that more aggressive lipid lowering was effective in preventing progression of atherosclerosis in SVGs, but warfarin had no effect. Using variables measured at baseline, we sought the independent prognostic factors for atherosclerosis progression in SVGs, employing the statistical method of generalized estimating equations with a logit-link function.
RESULTS
Twelve independent prognostic factors for atherosclerosis progression were found. In the order of their importance they were: maximum stenosis of the graft at baseline angiography; years post-SVG placement; the moderate low-density lipoprotein–cholesterol (LDL-C) lowering strategy; prior myocardial infarction; high triglyceride level; small minimum graft diameter; low high-density lipoprotein–cholesterol (HDL-C); high LDL-C; high mean arterial pressure; low ejection fraction; male gender; and current smoking.
CONCLUSIONS
This study identified Post-CABG patient and SVG characteristics associated with saphenous vein graft atherosclerosis progression. These data provide a basis for rational risk factor management to prevent progression of SVG atherosclerosis.
Aortocoronary saphenous vein bypass grafts (SVGs) are effective in the relief of angina and, in certain patient subsets, in prolonging life (1,2). However, the development and progression of atherosclerosis limits the duration of SVG usefulness. About 15% of SVGs are closed within 1 year (3–5). Between one and six years, the annual graft attrition rate is 1% to 2% and becomes 4% to 6% per year after that, so that about half of SVGs have significant stenosis or are occluded after 10 years. The development of vein graft atherosclerosis results in recurrent angina and the acute coronary syndromes. By 10 years, nearly one-third of patients undergo percutaneous or operative intervention (6). Second procedures carry more risk and are less likely to be successful than the initial operation (7).
Most of the stenosis and occlusion after the first year is caused by atherosclerosis. Prior studies have suggested that a number of prognostic factors for atherosclerosis progression in native arteries are operative in SVGs, including hyperlipidemia (7). The recently completed Post-Coronary Artery Bypass Graft (Post-CABG) trial demonstrated that aggressive, rather than moderate, treatment to lower serum cholesterol reduced significant progression of atherosclerosis in SVGs, whereas low-dose (1 to 4 mg daily) warfarin had no effect (8).
Attempts to improve treatment to prevent progression of SVG atherosclerosis would benefit from an improved understanding of definable prognostic factors and their relative order of importance. To this end, we studied the well-characterized patients entered into the Post-CABG trial (8) to determine prognostic factors associated with progression of atherosclerosis in SVGs and determined the relative contribution of these prognostic factors.
Methods
Post-CABG trial study design
The design of Post-CABG has been described previously (8). Briefly, patients 1 to 11 years’ postcoronary bypass surgery who met eligibility criteria were randomized in a 2 × 2 factorial design to moderate lipid lowering or more aggressive lipid lowering with lovastatin and, when necessary, to cholestyramine as well as to low-dose warfarin or placebo at baseline angiography. Each SVG was divided into three segments. Percent stenosis was determined as (1 − minimum graft diameter)/(maximum graft diameter in the corresponding graft segment) × 100. This means that the minimum graft diameter was not constrained to occur at the site of maximum stenosis. The baseline angiogram had at least two patent SVGs with stenosis <75% in men and at least one patent SVG in women; all patients had ejection fraction ≥0.30. All subjects were aged 21 to 74 years, and all had low-density lipoprotein cholesterol (LDL-C) levels 130 to 175 mg/dl and triglyceride levels below 300 mg/dl. In all, 1,351 patients were entered into the study; 92% were men and 94% white. The mean age was 61.5 years.
Follow-up angiograms were performed four to five years after enrollment to assess atherosclerotic progression. The primary end point of the study was the per-patient percentage of initially patent major grafts that had graft progression of atherosclerosis (“graft worsening”—that is, a decrease of ≥0.6 mm in lumen diameter) at the site of greatest change at follow-up.
The goal of this secondary analysis of the Post-CABG data was to determine prognostic factors for graft worsening.
Statistical analysis
Baseline measurements of potential prognostic value for significant graft worsening were selected by the investigators and are listed in Tables 1 and 2. ⇓⇓Continuous baseline measurements were compared in the two LDL-C lowering strategies using Satterthwaite’s test for the equality of means without equal variances (9). Dichotomous baseline variables (yes or no, or presence or absence of a characteristic) were compared using the Fisher exact test for the equality of proportions (10).
Summary of Continuous/Ordinal Baseline Measurements by Lipid-Lowering Treatment Assignment∗
Summary of Dichotomous Baseline Measurements by Lipid-Lowering Treatment Assignment∗
A logit-link function (10) was chosen to relate the dichotomous response variable of significant graft worsening to the potential prognostic (predictor) baseline variables listed in Tables 1 and 2. Let p denote the probability of significant worsening of a graft. The logit-link function models the natural log odds of p, g(p) = loge(p/[1 − p]), as a linear combination of an intercept (β0), selected variables (xi), and their corresponding parameter coefficients (βi), i.e.,
The β-coefficients were estimated from the data set using generalized estimating equations (GEE) (11), using a Statistical Analysis System (SAS) module written by Dr. Margaret M. Frederick (Clinical Trials & Surveys Corp.). This module obtains initial parameter estimates for a given model using the generalized linear model (GLM) and then iteratively solves the GEE equations to obtain the final GEE estimates. The GEE method takes into account the correlation among grafts within a subject and allows for each subject to have a different number of grafts. An unstructured correlation matrix was chosen, although the GEE method is robust to correlation matrix misspecification. All computations were performed in the SAS 6.12 programming language.
Because software for automated stepwise GEE procedures was not available, a modified version of backward elimination was used to select the best model. First, all variables were entered into the model, and those with p values greater than a threshold of 0.75 were eliminated. This step was repeated with thresholds of 0.5, 0.4, 0.3, 0.1, and 0.05. In a modified version of a forward step, variables with p values between 0.05 and 0.3 were individually added to a model that included all variables previously significant at the 5% level to see whether the newly added variable contributed further.
Highly correlated or colinear variables were not included in the same model. For example, separate models were constructed with (a) total cholesterol and (b) the three cholesterol components: high-density lipoprotein (HDL), LDL, and triglycerides, and then the model with better overall properties was chosen. Also, because a number of patients were missing some angiographic measurements, separate GEE models were constructed with and without the angiographic variables.
Because there do not exist well-established goodness-of-fit tests for GEE models (12), the goodness-of-fit statistics from the GLM (e.g., deviance, G2) were used as a surrogate. This choice was justified by the fact that the GEE and GLM methods give close point estimates and estimated variances.
To determine the relative significance of the variables included in the final model, a forward selection method was used to reconstruct the final model, with changes in the deviance (G2) used to assess improvement in fit of the model as variables were added. Each of the subsequent reductions in deviance (ΔG2) has approximately a chi-square distribution with 1 degree of freedom. This determined an ordering of the final prognostic variables from most to least important, as assessed by deviance.
Results
Tables 1 and 2 show the variables that were considered as potential prognostic factors in these analyses. Baseline lipid levels and blood pressures were those most recently obtained prior to enrollment. Patient self-report determined use of current medications. A question on antithrombotic use was added after the first 84 patients were enrolled, and thus information was incomplete on this variable.
Table 1 shows the means and standard deviation (SD) for the selected continuous variables at baseline for the two lipid-lowering treatment groups. These variables had approximately the same means in the two strategies, with two exceptions. The baseline mean body mass index (BMI) for the moderate group was lower than that of the aggressive group (p = 0.083) and the mean baseline pulse pressure was significantly lower in the moderate group (p = 0.044). Nevertheless, there was substantial overlap between the sample distributions of these variables, so they were not considered unbalanced.
Table 2 shows the numbers of observations and the proportions for “yes” outcomes of the selected dichotomous variables at baseline for the two lipid-lowering treatment groups. There was balance for these variables, except that the proportion of patients in the aggressive group using calcium channel blockers was greater than in the moderate group (20% for the moderate group vs. 27% for the aggressive group, p = 0.007).
Thus, it appears that the moderate treatment group might have had a small advantage in baseline factors over the aggressive treatment group. No warfarin versus placebo comparisons were considered because no warfarin effect was found in the Post-CABG trial (8).
Of the 1,351 patients enrolled in the Post-CABG trial, one was not eligible because there was at least 75% stenosis in all grafts, and 94 patients had missing outcome variables. Of the remaining 1,256 patients, some had missing baseline patient or subject–graft data: 8 patients had missing ejection fractions, including one who also had no subject–graft measurements for minimum diameter but had complete subject–graft measurements for maximum stenosis, leaving 1,248 patients. Of these, 1,105 patients had complete angiographic measurements; 98 additional patients had no subject-graft measurements for minimum diameter, of whom 10 had no, 19 had some (but not all), and 69 had all subject-graft measurements for maximum stenosis; and 45 patients had some subject-graft measurements for minimum diameter, of whom 25 had some and 20 had complete subject-graft measurements for maximum stenosis. These missing data reduced the total number of subject-graft measurements to 2,663 (in 1,248 patients) and 2,422 (in 1105 patients) without and with the inclusion of the angiographic variables, respectively. For the GEE analysis, missing measurements were regarded as occurring at random, and thus patients with missing responses or ejection fractions and subject–grafts with missing angiographic measurements were deleted. In addition, one patient had a missing history of coronary heart disease (CHD), and 84 patients had missing antithrombotic drug use because this variable was added after the study began. Separate analyses were performed with and without these variables, which did not enter the final model.
Table 3shows the 12 predictors in the final GEE model. For each predictor variable, the parameter estimates, estimated standard errors, p values, and odds ratios (ORs) with 95% confidence intervals (CIs) are shown. The predictor variables are listed in the order that they entered the GLM. Thus, all of these variables contributed a significant reduction in deviance over the variables that preceded them in the GLM estimation. The p value for the reduction in deviance is also shown. Thus, corresponding to equation (2), the estimated equation for the probability of significant graft worsening as a function of these prognostic variables is:
Estimates of Parameters, Standard Errors, and Odds Ratios for GEE Model for Graft Worsening as a Function of Baseline Variables
A positive coefficient indicates a direct relationship with graft worsening. Thus, the independent prognostic factors that make graft worsening more likely (in order of decreasing reduction in deviance) are: greater maximum stenosis of a graft; more years post-CABG; assignment to moderate (rather than aggressive) LDL-C lowering; prior myocardial infarction (MI); high triglycerides; smaller minimum diameter of the graft; low HDL-C; high LDL-C; high mean arterial pressure; low ejection fraction; being male rather than female; and being a current smoker.
Let β denote an estimated β-coefficient and X denote a predictor variable. Then exp (Xβ) gives the OR for the corresponding predictor variable when all other predictor variables are fixed. For a dichotomous variables, exp (β) indicates how much greater the odds of significant graft worsening are for a subject with a “yes” outcome on that predictor variable than for a subject with a “no” outcome on that predictor variable. For example, for a subject with a prior MI compared to a subject without a prior MI, the odds of significant overall graft worsening are 1.47 times larger, with a 95% CI of (1.17, 1.85).
The ORs for the continuous variables indicate how much greater the odds of significant graft worsening increase for a subject with a 1-unit increase in the predictor variable. For example, for a 10 mm Hg increase in mean arterial pressure, the odds of significant graft worsening are increased by exp(10∗0.0126) = 1.13 times, with a 95% CI of (1.02, 1.26).
Table 4shows the results of the GEE model with the same predictor variables as in Table 3, except for the angiographic variables, which were omitted. It is shown because 98 additional subjects are included in the model. Although the parameter estimates and estimated variances are close to those in Table 3, it is clear that the inclusion of the angiographic variables and the deletion of the 98 patients with some missing data does affect the other parameter estimates. In particular, the intercept shows the greatest change, followed by the dichotomous variable for female.
Estimates of Parameters, Standard Errors, and Odds Ratios for GEE Model for Graft Worsening as a Function of Baseline Variables (no angiographic variables)
Discussion
This study defined the prognostic factors for risk of atherosclerotic progression in SVGs in the large and well-characterized Post-CABG patient population. In addition, the relative importance of the prognostic factors (in terms of reduction in deviance) was ascertained. These factors (in order of their reduction in deviance) were maximum percent stenosis in the graft at baseline angiography; years since SVG placement; moderate (per the Post-CABG protocol) treatment of LDL cholesterol; a history of prior MI; elevated triglyceride level; small minimum graft diameter; low HDL and high LDL cholesterol levels; high mean arterial pressure; low ejection fraction; male gender; and current smoking. The model permits one to calculate the probability of atherosclerosis progression in grafts in this patient cohort and to assess quantitatively the relative importance of each of the variables.
Comparison with prior studies
Campeau et al. (13) reported that the treatment effect of aggressive compared with moderate LDL-C lowering in the Post-CABG trial did not differ significantly among subgroups of age (<60 or >60 years), gender, or CHD risk factors such as smoking, hypertension, diabetes mellitus, HDL <35 mg/dl versus ≥35 mg/dl, and tryglyceride levels <200 mg/dl compared to ≥200 mg/dl. They found that the change in minimum lumen diameter was in the same direction for all subgroup categories and there were no significant interactions with treatment.
Here we found a significant association of graft age and progression of atherosclerosis. True atherosclerosis, defined by the World Health Organization as the presence of foam cells (14), has never been observed before the first postoperative year and seldom before the fourth year (15,16). Several studies involving two to four repeat angiographic examinations have reported a progressive increase in late obstructive changes with increasing time after the initial operation. In one study of 98 grafts in 82 patients that appeared normal one year after surgery, late obstructive changes were observed in 24 grafts at five to seven years and in 62 between 10 and 12 years (17). Cosgrove et al. (18) studied 665 patients who underwent repeat coronary bypass because of graft failure and/or progression of disease in the native coronary arteries. They reported an incidence of reoperation of 2.7% at 5 years, 11.4% at 10 years, and 17.3% at 12 years following initial grafting. Solymoss et al. (19) studied 119 consecutive patients who had symptom-directed angiography 17 to 203 (mean 95 ± 5) months after coronary bypass. They observed late graft changes compatible with atherosclerosis or thrombosis in 70% of patients. A stepwise logistic regression analysis identified graft age as an independent risk factor for angiographic progression.
Consistent with the results of the Post-CABG study (8), substantial data are present in the literature to support a role for elevated lipid levels in the progression of SVG atherosclerosis. Campeau et al. (3) observed that elevated plasma cholesterol and LDL as well as low HDL were associated with atherosclerotic progression in grafts. Daida et al. (20) found a strong relationship of cholesterol level and vein graft obstruction in 284 patients a mean of seven years after bypass. Stenosis ≥70% was present in 12% of grafts in patients with serum cholesterol <200 mg/dl, 39% in patients with serum cholesterol 200 to 239 mg/dl and 43% in patients with serum cholesterol ≥240 mg/dl. Additionally, elevated serum lipid levels, including elevated triglyceride levels, have been associated with late clinical events following bypass surgery, including MI and need for further revascularization, which may have resulted from graft failure associated with atherosclerosis (21,22). Neitzel et al. (23) found atherosclerosis in 71% of the 50 grafts in patients who had reoperation or died six to 12 years after their first bypass operation. The prevalence of coronary artery disease risk factors in this cohort was compared to a control population of 535 survivors who did not require another revascularization procedure during the same period. Higher total cholesterol and triglyceride levels (p < 0.001) and lower high-density lipoprotein levels (p < 0.05) were found in the serum of patients who required a second bypass operation than in the controls. The stepwise logistic regression analysis of Solymoss et al. (20) identified apolipoprotein B and Lp(a) levels as independent risk factors for progression.
A number of other studies have also identified Lp(a) as predictive of SVG atherosclerosis (3,24,25). In an autopsy study, Lie et al. (15) found atherosclerosis in 3 of 26 (11.5%) grafts from normolipemic patients and in 11 of 14 (78.6%) grafts from hyperlipemic patients who survived 13 to 75 months following surgery. Atkinson et al. (26) did an autopsy study on 56 patients 12 to 168 months following coronary bypass. Hypercholesterolemia was present in 68% of patients in whom graft atherosclerosis was found but only in 15% of patients having grafts with only intimal hyperplasia (p < 0.02).
In this study, we found that maximum graft stenosis and, also, minimum graft diameter at baseline angiography were predictive of angiographic progression between baseline and follow-up angiography. Most, but not all, of the available data are consistent with our findings. Campos et al. (27) presented data on long-term follow-up of normal and minimally diseased (<35% stenosis) SVGs. This group studied 62 patients with a postoperative angiogram showing normal or minimally diseased SVGs 6.1 ± 2.1 years after their initial bypass operation. A repeat angiogram was performed 5.1 ± 1.4 years later and this was used to assess graft progression. More than half of these grafts remained normal or minimally diseased. Data from Mehta et al. (28) also suggest a favorable outcome if SVGs are normal at baseline. Kouz et al. (29) in a study of 144 patients one year and 8 ± 5 years post-SVG placement reported a positive correlation between early graft narrowing and late graft occlusion.
In our study, higher mean arterial pressure was associated with progression of SVG atherosclerosis. In the study of Campos et al. (27), severe hypertension was identified as a risk factor for atherosclerotic progression.
The study of Cosgrove et al. (18) found, as did the present study, that left ventricular dysfunction also predicted progression of vein graft atherosclerosis.
In Post-CABG, female gender was associated with reduced SVG atherosclerotic progression, although this was one of the lesser risk factors. The investigators are aware of no other data that speak to this issue.
Current smoking was significantly associated with SVG atherosclerosis progression in Post-CABG patients, though it appears to be one of the lesser risk factors from a quantitative point of view. This is consistent with data from studies that have shown angiographic progression of SVG atherosclerosis in smokers compared to nonsmokers (30) and graft thrombosis (31).
There was no effect of baseline therapy with antithrombotic agents in this study. In a randomized trial reported by Chesebro et al. (32), the combination of dipyridamole and aspirin was compared to placebo for the reduction of SVG occlusion. At one month and one year a significant reduction occurred in graft occlusion as a result of treatment with aspirin and dypyridamole. These two studies are not directly comparable, however, because we had very few occlusions and the Chesebro study was based entirely on occlusion.
Clinical implications
An understanding of the prognostic factors that lead to the progression of SVG atherosclerosis provides a rational strategy for designing therapeutic interventions. In addition to showing that more intensive lipid lowering is important (this was also the main conclusion of the Post-CABG trial), our data suggest that smoking cessation should reduce atherosclerosis progression in SVGs.
The main results of the Post-CABG trial (8) and those of the Cholesterol-Lowering Atherosclerosis Study (CLAS I and II) (33,34) indicate the importance of lipid lowering. In addition, these data suggest the importance of blood pressure control. Also, a reasonable hypothesis arising from this study is that reduction of triglycerides may be effective in preventing SVG atherosclerotic progression.
Study limitations
This study was a secondary analysis of the Post-CABG clinical trial data. Because patients in this study were accrued between 1989 and 1991, certain candidate prognostic factors, including fibrinogen and apolipoprotein A, were not studied in all patients. Further, the analysis presented herein should be considered to be exploratory because many analyses were undertaken to derive the final model. The equation derived to calculate the probability of atherosclerosis progression is strictly applicable only to the Post-CABG population and should be validated on an independent data set. However, it seems likely that it would apply qualitatively to other coronary populations with similar baseline characteristics. The degree of baseline atherosclerotic disease no doubt varied in this population, which was studied one to 11 years following initial surgery, and the grafts in these patients were likely at different points in the development of their atherosclerotic disease. For this reason, the time interval post-CABG was incorporated into the multivariate analysis.
Conclusions
This study identified Post-CABG patient and graft characteristics associated with SVG atherosclerosis progression including: maximum stenosis of the graft at baseline angiography; years post-SVG placement; the moderate LDL-C lowering strategy; prior MI; high triglyceride level; minimum graft diameter; low HDL cholesterol; high LDL cholesterol; high mean arterial pressure; low ejection fraction; male gender; and current smoking. These data provide a basis for rational risk factor management to prevent progression of SVG atherosclerosis.
Footnotes
↵# The Post-CABG Trial Investigators are listed in reference .
☆ Dr. Borkowf was supported by a National Research Council postdoctoral fellowship.
- Abbreviations
- CHD
- coronary heart disease
- GEE
- generalized estimating equation
- GLM
- generalized linear model
- HDL-C
- high-density lipoprotein cholesterol
- LDL-C
- low-density lipoprotein cholesterol
- Post-CABG
- Post-Coronary Artery Bypass Graft trial
- SAS
- Statistical Analysis System
- SVG
- saphenous vein graft
- Received February 28, 2000.
- Revision received June 22, 2000.
- Accepted July 31, 2000.
- American College of Cardiology
References
- ↵
- Davis K.,
- Chaitman B.,
- Ryan T.,
- Bittner V.,
- Kennedy J.
- ↵
- Campeau L.,
- Enjalbert M.,
- Lesperance J.,
- et al.
- Bourassa M.
- Fitzgibbon G.,
- Kafka H.,
- Leach A.,
- Keon W.,
- Hooper D.,
- Burton J.
- ↵
- ↵
- Motwani J.,
- Topol E.
- ↵
- ↵
- Gaylor D.W.
- ↵
- Agresti A.
- ↵
- Liang K.-Y.,
- Zeger S.L.
- ↵
- ↵
- Campeau L.,
- Hunninghake D.B.,
- Knatterud G.L.,
- et al.
- ↵WHO classification of atherosclerotic lesions. Report of a Study Group. Tech Rep Ser 1958;143:33.
- ↵
- Solymoss B, Leung T, Pelletier L, Campeau L. Pathologic changes in coronary artery saphenous vein grafts and related etiologies. In: Waters D, Bourassa M, editors. Care of the Patient With Previous Bypass Surgery. Cardiovascular Clinics, 1991, 45–65.
- ↵
- Lawrie G.,
- Morris G.,
- Earle N.
- ↵
- ↵
- ↵
- Daida H.,
- Yokoi H.,
- Miyano H.,
- et al.
- ↵
- Van Brussel B.,
- Plokker T.,
- Ernst S.,
- et al.
- ↵
- Campeau L, Enjalbert M, Lesperance J, et al. Atherosclerosis and late closure of aortocoronary saphenous vein grafts: sequential angiographic studies at 2 weeks, 1 year, 5 to 7 years, and 10 to 12 years after surgery. Circulation 1983;Suppl II:68II1.
- Hoff H.,
- Beck G.,
- Skibinski C.,
- et al.
- ↵
- Atkinson J.,
- Forman M.,
- Vaughn W.,
- et al.
- ↵
- Campos E.,
- Cinderella J.,
- Farhi E.
- ↵
- Mehta I.,
- Zaret B.,
- Weinberg J.,
- Kopf G.,
- Elefteriades J.
- ↵
- Kouz S.,
- Campeau L.,
- Lesperance L.,
- et al.
- ↵
- ↵Solymoss B, Nadeau P, Millette D, Capeau L. Late thrombosis of saphenous vein coronary bypass grafts related to risk factors. Circulation 1988;78Suppl I:I-140–3.
- ↵
- ↵