HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Add to Personal Folders Download to citation manager Author home page(s): Bruce W. Lytle Eugene H. Blackstone Floyd D. Loop John H. Arnold Rami Akhrass Patrick M. McCarthy Delos M. Cosgrove Permission Requests Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Lytle, B. W. Articles by Cosgrove, D. M. Search for Related Content PubMed PubMed Citation Articles by Lytle, B. W. Articles by Cosgrove, D. M.

J Thorac Cardiovasc Surg 1999;117:855-872
© 1999 Mosby, Inc.

SURGERY FOR ADULT CARDIOVASCULAR DISEASE

TWO INTERNAL THORACIC ARTERY GRAFTS ARE BETTER THAN ONE

From the Department of Thoracic and Cardiovascular Surgery, The Cleveland Clinic Foundation, Cleveland, Ohio.

Read at the Seventy-eighth Annual Meeting of The American Association for Thoracic Surgery, Boston, Mass, May 3-6, 1998.

Received for publication May 8, 1998. Revisions requested June 10, 1998. Revisions received Jan 21, 1999. Accepted for publication Jan 22, 1999. Address for reprints: Bruce W. Lytle, MD, The Cleveland Clinic Foundation, F25, 9500 Euclid Ave, Cleveland, OH 44195. J Thorac Cardiovasc Surg 1999;117:855-72

	Abstract

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 
Objective: Does the use of bilateral internal thoracic artery (ITA) grafts provide incremental benefit relative to the use of a single ITA graft?
Methods: We conducted a retrospective, nonrandomized, long-term (mean follow-up interval of 10 postoperative years) study of patients undergoing elective primary isolated coronary bypass surgery who received either single (8123 patients) or bilateral ITA grafts (2001 patients), with or without additional vein grafts. Multiple statistical methods including propensity score matching, and multivariable parsimonious and nonparsimonious risk factor analyses were used to address the issues of patient selection and heterogeneity.
Results: In-hospital mortality was 0.7% for both the bilateral and single ITA groups. Survival for the bilateral ITA group was 94%, 84%, and 67%, and for the single ITA group 92%, 79%, and 64% at 5, 10, and 15 postoperative years, respectively (P < .001). Death, reoperation, and percutaneous transluminal coronary angioplasty were more frequent for patients undergoing single rather than bilateral ITA grafting, and this observation remained true despite multiple adjustments for patient selection, sampling, and length of follow-up. The differences between the bilateral and single ITA groups were greatest in regard to reoperation. The extent of benefit of bilateral ITA grafting varied according to patient-related variables, but no patient subsets were identified for whom single ITA grafting could be predicted to provide an advantage.
Conclusions: Patients who received 2 ITA grafts had decreased risks of death, reoperation, and angioplasty. (J Thorac Cardiovasc Surg 1999;117:855-72)

	Introduction

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 
The strategy of using the left internal thoracic artery (LITA) as a graft to the left anterior descending (LAD) coronary artery has become a standard part of coronary bypass surgery. LITA-LAD grafts have high long-term patency rates, graft atherosclerosis is rare, and clinical studies have demonstrated improved long-term survival and fewer repeat procedures and cardiac events for patients receiving LITA-LAD grafts. ^1-4 Logic seems to dictate that adding the use of the right internal thoracic artery (RITA) as a bypass graft might further improve long-term outcomes. However, no studies have clearly demonstrated incremental benefits of bilateral ITA grafting over those achieved with an LITA-LAD graft combined with saphenous vein grafts. ^5-15

There are many possible reasons that it has been difficult to show a clinical advantage for the use of 2 ITA grafts, including (1) the success of the single ITA graft in achieving good outcomes for at least a decade, requiring relatively long-term follow-up in a large number of patients to demonstrate incremental benefit of multiple ITA grafts; (2) the multiple factors leading to the selection of patients to receive bilateral versus single ITA grafts confounds the interpretation of apparent benefits.

In this observational, nonrandomized study, we have compared outcomes for a large group of patients whose primary elective bypass surgery included either the use of 1 or both ITAs. The methods of comparison addressed pitfalls of using clinical experience to draw a casual inference that include (1) patient selection, potentially confounding the comparison, (2) patient heterogeneity influencing the overall estimate of any benefit because of patient mix, and (3) over-statement of the magnitude and certainty of any benefit contributed in part by the statistical methods used, as well as from the use of a single experience. The outcome end points that were studied were death from any cause, cardiac reoperation for any reason, and percutaneous treatment of coronary artery disease (percutaneous transluminal coronary angioplasty, or PTCA).

	Patients and methods

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 
Patient population.
With the aid of a computerized prospective cardiovascular information registry, we identified all patients undergoing primary isolated revascularization using both ITAs for bypass grafts during the years 1971 to 1989. Those patients are termed the bilateral ITA group. The first bilateral ITA operation at The Cleveland Clinic Foundation was performed in 1971. Entry to the study was stopped in 1989 to provide a minimum 6-year follow-up interval for all patients. Foreign patients were excluded from this group because of anticipated difficulty in follow-up. From the group of 2015 bilateral ITA patients identified, 14 patients were undergoing emergency operation and were excluded from further study because emergency operation is a factor that strongly affects the selection of patients for arterial graft operations. Therefore, aside from the exclusion of foreign patients and the 14 patients undergoing emergency operation, this represents a consecutive series of patients receiving bilateral ITA grafts as a primary operation for isolated bypass grafting.

In addition, through our cardiovascular information registry we have established a database of preoperative and perioperative data and have periodically conducted follow-up studies involving the first 1000 patients per year undergoing primary isolated elective (emergency operations excluded) coronary artery bypass grafting. From those 1000 patient cohorts undergoing surgery during the years 1971 to 1989, all patients were identified who received 1 ITA graft and at least 1 saphenous vein–coronary artery bypass graft. There were 8123 such patients, termed the single ITA group. Some foreign patients were included in these cohorts. Preoperative characteristics of both patient groups are listed in Appendix 1. No aspect of the surgical process was randomized for these patients. Multiple strategies for myocardial protection were used. The decision to use single or bilateral ITA grafts and the vessels that were chosen to be grafted with those ITAs were at the discretion of the attending surgeon.

In the bilateral ITA group, 1693 patients (84.6%) received 2 ITA grafts, 292 (14.6%) received 3, and 16 (0.8%) received 4. The ITAs were used entirely as in situ grafts for 1744 (87.2%) patients, 248 (12.4%) patients received 1 free ITA graft, and 9 received 2 free ITA grafts. Free grafts were performed as aorta-coronary grafts. Major target vessels for the ITA grafts were LAD and circumflex, 1217 patients (60.8%); LAD and right coronary artery (RCA), 459 patients (22.9%); and circumflex and RCA, 58 patients (3.0%). The rest of the patients (13.3%) received grafts that involved at least 1 ITA graft being used to graft a vessel classed as a diagonal coronary artery. Ramus, or intermediate arteries, fell into this classification. The details of the ITA grafting strategies are listed in Appendix 2. Nineteen patients (0.9%) who received concomitant gastroepiploic grafts and 4 (0.2%) who received radial artery grafts were excluded from the multivariable analyses at the request of one of the reviewers of this manuscript. In the single ITA group the ITA was used as a graft to the LAD in 7147 patients, the LAD-diagonal in 144 patients, the circumflex in 220 patients, the diagonal branch in 588 patients, the RCA in 22 patients, and the diagonal-circumflex arteries in 2 patients. The patient numbers and percentages according to the number of saphenous vein grafts received were 0 saphenous grafts, 436 patients (21.8%); 1 graft, 750 patients (37.5%); 2 grafts, 599 patients (29.9%), 3 grafts, 182 patients (9.1%), 4 grafts, 31 patients (1.5%), and 5 saphenous grafts, 3 patients (0.1%).

Routine follow-up by mail with supplemental telephone follow-up has been conducted every 5 years for the single ITA group. Patients in the bilateral ITA group were followed up within 1 year of data analysis and had received at least 1 previous follow-up. Because of our interest in multiple arterial grafting, most patients in the bilateral ITA group had been followed up more than twice. The median follow-up interval of hospital survivors was 10.13 years (mean 9.71 ± 3.0 years) for the bilateral ITA group and 10.26 years (mean 10.80 ± 5.2 years) for the single ITA group. In the bilateral ITA group, follow-up was incomplete for 51 patients. Of these 51 patients, 6 had 1 to 5 years of postoperative follow-up, 34, 5 to 10 years, and 5 more than 10 years. In the single ITA group, 634 patients had incomplete follow-up, with 82 followed up for 1 to 5 years, 198 for 5 to 10 years, and 189 more than 10 years. The end points that were recorded were death from any cause, cardiac reoperation for any reason, and attempted percutaneous intervention for coronary artery disease (PTCA). When patients had incomplete follow-up they were censored at the date of last follow-up.

Logistic regression analysis was used to determine important predictors of in-hospital mortality. A logistic regression model, adjusted for age and ITA group, was then fit with only those variables that were significant in the stepwise model (P < .05) to allow the maximum number of patients to be analyzed. The odds ratios and 95% confidence intervals have been provided (Table I).

Nonparametric estimates of survival, freedom from cardiac reoperation, and freedom from PTCA were estimated for each group by the method of Kaplan and Meier. ¹⁶ The instantaneous risk of events across time (the hazard function) was estimated by means of a parametric method that resolved the number of hazard phases, identified the shape of the hazard function, and estimated its parameters. ¹⁷

Preliminary data analysis
Preliminary exploration of the data with respect to each event considered included frequency data of preoperative baseline characteristics using the ² test to detect differences unlikely to be due to chance between the 2 ITA groups. Mean values of continuous variables, such as age, were compared by means of the Student t test. Correlations were sought between variables. Life tables stratified by categoric variables were explored. Because naturally ordered variables, such as anginal symptoms and degree of left ventricular function, and all continuous variables were retained as such in the analyses, we investigated multiple transformations of these variables by decile risk analysis to best calibrate them in the risk factor models. These preliminary analyses were repeated in the case of hazard function analyses for events subdivided according to the time phase of their occurrence.

Data analysis
The primary objective of the data analysis was the statistical estimation of the possible benefit of bilateral ITA grafting on survival and freedom from interventions. Other objectives included the assessment of the effect of patient-related variables on outcome and the identification of definable patient subgroups with particularly great or particularly little benefit from bilateral ITA grafting. The analysis of possible benefit addressed confounding (patient selection) by use of a propensity score, ¹⁸ heterogeneity (risk factors) by multivariable risk factor analysis, ¹⁹ and shrinkage (exaggeration of benefit) by bootstrap resampling. ²⁰

Confounding (patient selection)
Because use of bilateral versus single ITA grafting did not involve randomized assignment, patient selection bias was addressed by a matching strategy for which a propensity score was used. ¹⁸ The event bilateral ITA (vs single ITA) grafting was analyzed, without regard to any outcome, using logistic multivariable regression. The object was not to develop a parsimonious risk factor model, but to account for all known sources of potential selection bias. Nevertheless, to accomplish this, a sequence of intermediate modeling efforts was needed. First, a parsimonious model of individual factors ("main effects") was developed independently by P.H. and E.B. and the models were reconciled. All factors listed in Appendix 3 were used, but unlike the subsequent multivariable analyses of outcomes, variables for individual surgeons were also incorporated, because they could be considered a source of selection bias. Second, interactions among the variables in the parsimonious model were sought, these being dominated by age of the patient and date of the operation. Finally, risk factors not represented in the parsimonious model were incorporated. These additional variables included anginal symptoms, history of myocardial infarction, family history of coronary artery disease, left main disease, hypertension, and chronic obstructive pulmonary disease. The area under the receiver operating characteristic (ROC) curve for this model was 0.91.

After completion of the propensity model, a propensity score was calculated from the logistic equation for each patient. The model predicted likelihoods of bilateral ITA grafting (vs single ITA grafting) ranging from essentially 0% to 99.7%. As suggested by Rubin, ²¹ the patients were sorted by propensity and compared within quintiles. Only 68 patients receiving bilateral ITA grafting were in the first 2 quintiles. However, bilateral and single ITA patients in quintiles 3 through 5 were well matched with respect to patient risk factors. Appendix 4 details the matching of quintiles 3 through 5 in regard to selected variables, but the bilateral ITA–single ITA groups were similarly well matched for all variables in Appendix 3. Thus the sample of "propensity-matched patients" excludes the first 2 quintiles of single ITA patients (3976 patients). The excluded patients were predicted to receive bilateral ITA grafting with low probability for any of a variety of reasons, and essentially did not match bilateral ITA grafting cases for comparison purposes. The propensity score was either retained as a logit unit score or transformed into the probability domain (as the probability of receiving bilateral ITA grafting). It was incorporated as a continuous variable into outcomes analyses described below to adjust for possible confounding. ²²

Heterogeneity (risk factors)
A number of risk factors (case mix) influence each of the outcomes studied; the prevalence of some of these differed between the groups (Appendix 1). These differences may have increased or reduced the apparent benefit of bilateral versus single ITA grafting. Importantly, these factors also could give insight into the nature of the treatment and predict which patients are more or less likely to benefit from treatment. Therefore risk factor models were developed for the separate events, death, reoperation, and PTCA.

Two multivariable risk factor models were used for 2 different purposes. Semiparametric Cox proportional hazards regression was used to obtain average effects across time. ²³ Interactions of variables with the type of ITA grafting were not considered to obtain a simple estimate of a possible benefit of bilateral ITA grafting in the form of a single parameter estimate. The second method used parametric, multiphase, nonproportional hazards regression to obtain time-specific effects. ¹⁷

For the Cox proportional hazards regression, initial screening of potential risk factors was accomplished by forward stepwise regression, followed by examination of interaction terms. For the nonproportional hazards regression, a directed technique of entry of variables into the multivariable risk factor model was used. ¹⁸Both forward stepwise and backward variable elimination analysis supplemented this directed technique, identifying the same model. Interactions among variables were sought thereafter, including interactions with type of grafting.

For the Cox proportional hazards model, results are presented in terms of the regression coefficients ± 1 standard error and the hazard ratio and 95% confidence limits. Nonproportional hazard regression coefficients are presented ± 1 standard error in a phase-specific fashion. The P value criterion for retention of variables and interactions in the final models was .05 for either method.

While these modeling efforts were performed quasi-independently, after the development of each, differences between models were examined, and the differences were either reconciled or understood as genuine differences in the models. Thus all final models represent the end product of a joint statistical collaborative effort.

Confounding and heterogeneity
To maximally adjust the estimate of possible benefit of bilateral ITA grafting, nonparsimonious regression models were developed in both the Cox proportional and the parametric nonproportional hazards domains with both the type of ITA grafting forced into the model, as well as the most appropriate transformation of the propensity score. Initially parsimonious models were developed using a P value criterion of .1, followed by investigation of interaction terms (including interactions with type of grafting in the case of the parametric nonproportional hazards modeling). Interactions were included only for P < .05. In general, the variables found in the parsimonious aspects of this modeling effort were identical to those in the analysis of heterogeneity. Subsequently, variables were added to the model, irrespective of P value, from each category of variables. Thus the final models always included gender, age (or its transformation), body size (height or weight, whichever was more significant), grade of anginal symptoms, degree of left ventricular dysfunction, presence of congestive heart failure, history of previous myocardial infarction, family history of coronary disease, smoking history, peripheral vascular disease, diabetes, hypertension, one or more variables concerning the extent of coronary artery disease, date of operation, and location of ITA grafts. In the case of the parametric nonproportional hazards model, the same procedure was followed for the hazard phase containing the largest number of events; however, in the case of phases containing few events, only that number of additional variables was added that was less than half the number of events identified for that phase.

Shrinkage (exaggeration of benefit)
All these analyses were performed on the same groups of patients. This has the disadvantage that the apparent benefit may be inflated and the variance estimate too small, resulting in P values for the difference that are smaller than might be found in repeated similar experiences. Various methods have been developed, therefore, to "shrink" the estimates, yielding a more conservative, but probably more accurate, estimate of benefit. In this study, multiple patient samples, each equal in number to the total propensity-matched sample size, were randomly drawn from the existing patient population. Each sample was then analyzed as though it were a new patient population and all coefficients of the proportional and nonproportional hazards models adjusted for confounding and heterogeneity. This sampling and analysis process was repeated 1000 times. This computer-intensive process called bootstrapping yielded the most conservative estimate of the effect of arterial grafting strategy. ²⁰

Data in figures are presented with confidence limits equivalent to 1 standard error. In tabular data means are presented ± 1 standard deviation.

	Results

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 
Patient characteristics
Appendix 1 shows the univariate comparisons of the baseline differences between the entire bilateral ITA and single ITA groups. There were statistically significant differences in regard to gender, age, incidence of diabetes, family history of coronary artery disease, preoperative functional class, left ventricular function status, extent of disease (number of vessels stenotic), and greater than 70% LAD lesion (all P < .001). All variable subgroups were well represented in both the bilateral and single ITA groups except that there were a small number of patients in the bilateral ITA group who were 70 years old or older (62 patients).

In-hospital morbidity and mortality
Hospital death occurred in 14 patients (0.7%) in the bilateral ITA group and 58 patients (0.7%) in the single ITA group. Table I contains the results of logistic regression analysis of the variables in Appendix 1 in regard to their influence on in-hospital mortality. The number of ITAs used as grafts was not associated with mortality. The only morbid event significantly different between the 2 groups was sternal wound complications (bilateral ITA 50/2001, 2.5%, vs single ITA 115/8123, 1.4%) (P = .001).

View larger version (20K): [in this window] [in a new window]   Fig. 1. Comparison of the bilateral ITA (BITA) and single ITA (SITA) groups in terms of survival (A) and reoperation-free survival (B). Numbers of patients surviving at selected follow-up intervals are listed beneath part A.

View larger version (28K): [in this window] [in a new window]   Fig. 2. Comparison of survival and reoperation hazard function curves in the propensity-matched patients (both P < .0001) (bilateral [BITA], n = 1989; single [SITA], n = 4147). CABG, Coronary artery bypass grafting.

Analyses of differences in outcomes
The simplest question of the study (Are 2 ITA grafts better than 1?) is addressed quantitatively in Table III by means of multiple statistical approaches. This table presents the hazard ratios from Cox proportional hazard regression models for each outcome (death, reoperation, PTCA) and includes (1) a nonadjusted estimate based on the entire bilateral ITA and single ITA population for all follow-up available, (2) a nonadjusted estimate for the entire population with follow-up truncated to 12 years (because of small numbers of bilateral ITA patients followed up for more than 12 years), (3) a nonadjusted estimate for the propensity-matched patients for 12 years, (4) estimates adjusted for confounding only in the propensity-matched groups (using the propensity score), (5) estimates based on traditional parsimonious modeling of the propensity-matched groups, (6) estimates based on nonparsimonious adjustments for both confounding and heterogeneity, and, finally, (7) estimates based on the 1000 nonparsimonious bootstrap models drawn from the propensity-matched groups adjusted for both confounding and heterogeneity. The benefit of bilateral ITA grafting is expressed by the hazard ratio whose value is equal to 1 if the average outcomes are equivalent and greater than 1 if outcomes are better with 2 ITA grafts. The last line in Table III produces the most conservative and believable set of estimates. However, regardless of the adjustment techniques used, the estimates are fairly similar and in all cases outcomes were better for the patients with 2 ITA grafts.

View larger version (28K): [in this window] [in a new window]   Fig. 3. Distribution of Cox hazard ratios for 1000 "bootstrap" samples tested for the influence of bilateral versus single ITA grafts on the occurrence of death, PTCA, and reoperation. Any Cox hazard ratio greater than 1 indicates an advantage for bilateral ITA grafting. All hazard ratios for all samples were greater than 1.

View larger version (31K): [in this window] [in a new window]   Fig. 4. Nonproportional hazard equations were solved twice for each propensity-matched patient, once as if the patient had single ITA grafting and once as if the patient had bilateral ITA grafting. The distribution of the percentage differences in outcomes for all patients based on bilateral versus single ITA grafting are shown in this figure. All patients had better predicted outcomes with bilateral ITA grafting. The percentage of the patient population is expressed vertically and the percent difference in the occurrence of each event is expressed horizontally. The table details percent differences at selected percentages of the entire patient population.

View larger version (30K): [in this window] [in a new window]   Fig. 5. Non-risk-adjusted age relation to survival and freedom from reoperation at 12 postoperative years for propensity-matched patients based on nonproportional hazard equations. The survival curve confidence limits of patients having bilateral (BITA) and single (SITA) ITA grafting overlap at young ages. Bilateral ITA grafting produces significantly better freedom from reoperation at all ages and better survival at older ages.

View larger version (19K): [in this window] [in a new window]   Fig. 6. Non-risk-adjusted survival curves for diabetic and nondiabetic patients according to bilateral (BITA) versus single (SITA) ITA grafting at 12 postoperative years. For both groups survival for patients having bilateral ITA grafting was better.

View larger version (21K): [in this window] [in a new window]   Fig. 7. Twelve-year survival of patients having bilateral (BITA) or single (SITA) ITA grafting according to left ventricular (LV) function (none or mild dysfunction vs moderate or severe dysfunction). For both ventricular function subgroups, survival was better for patients having bilateral ITA grafting.

For illustration we used nonproportional hazard equations (Tables IV and V) to examine the difference based on bilateral versus single ITA grafting for patients defined by multiple characteristics. Fig. 8 depicts predicted outcomes for patient 1 (a 47-year-old normotensive man with normal left ventricular function, class II angina, and 2-vessel disease) and patient 2 (a 64-year-old hypertensive man with a previous myocardial infarction, 3-vessel disease, and moderate impairment of left ventricular function). Patient 1 is predicted to have excellent survival regardless of whether he receives 1 or 2 ITA grafts, but he has a much higher likelihood of reoperation (approximately 10% ¥ 12 postoperative years) if he receives a single ITA graft. Patient 2 has a worse survival outlook overall, but the risks of death and reoperation are both significantly decreased by the use of 2 ITA grafts. Age, however, is not the only important predictor. Fig. 9 depicts outcomes for 2 patients with 3-vessel disease and class III angina. Patient 3 is a 70-year-old man with normal left ventricular function and patient 4 is 50 years old with severe left ventricular dysfunction, a history of hypertension, and heart failure. Both patients are predicted to have improved survival with bilateral ITA grafting (patient 4 more than patient 3), and although both patients are also at a decreased risk of reoperation with bilateral ITA grafting, again, patient 4 benefits more than patient 3.

View larger version (28K): [in this window] [in a new window]   Fig. 8. Predicted survival and reoperation hazard function curves according to 1 ITA versus 2 ITAs for a 47-year-old normotensive man with 2-vessel disease and normal left ventricular function (patient 1) and for a 64-year-old hypertensive man with 3-vessel disease and moderate impairment of left ventricular function (patient 2). CABG, Coronary artery bypass grafting.

View larger version (33K): [in this window] [in a new window]   Fig. 9. Predicted survival and reoperation hazard function curves, 1 ITA versus 2 ITAs, for a normotensive 70-year-old man with normal left ventricular function (patient 3) and a 50-year-old hypertensive man with severe left ventricular dysfunction and a history of congestive heart failure (patient 4).

	Discussion

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

A major issue in the investigation of the relative benefits of single versus bilateral ITA grafting is patient selection. Even advocates of bilateral ITA grafting have not used this strategy for all patients, and the possibility that differences in long-term outcomes are based on patient-related variables or the period of the operation rather than operative strategy must be realistically considered. Our bias against using bilateral ITA grafts during emergency operations during the period of this study was so strong that we excluded patients undergoing emergency surgery from analysis. Diabetic patients and women were less likely to receive 2 ITA grafts, but the differences were small (diabetic: bilateral ITA 11%, single ITA 13%; female: bilateral ITA 11%, single ITA 16%). Advanced age also had a negative correlation with the use of bilateral ITA grafts, based mainly on smaller numbers of patients over 70 years of age in the bilateral ITA group. Also, the use of 2 ITA grafts was not consistent throughout the time period of the study. Until approximately 1980 the use of 2 ITA grafts was limited to situations in which venous conduits were not available or for extremely young patients. Thus some patients in the bilateral ITA group have been followed up for more than 20 years, but the median follow-up of the bilateral ITA group was just over 10 years, and fewer than 100 patients were followed for more than 13 years. Conclusions from this study are most valid for a 12-year time frame.

The ideal way to eliminate bias in the selection of treatment is to conduct a randomized clinical trial. Randomization should neutralize selection bias and allow comparison of treatment strategies for equal patient subsets. However, substantial difficulties are associated with the concept of a randomized trial comparing single ITA versus bilateral ITA grafting. The experience of the Coronary Artery Surgery Study (CASS) and the Bypass Surgery Angioplasty Revascularization Investigation (BARI) has taught us that selection bias is profound in randomized trials, but that selection bias is exerted at the point of inclusion into the trial rather than at the point of treatment choice. Expense is a major issue associated with these large trials, and even if funding and participants were available today, 10-year results would not be available for at least 15 years.

The standard of randomized clinical trials can be achieved only partially in nonrandomized comparisons. Coupling comparisons with traditional risk factor analysis adjusts for at least some factors related to selection so long as they also influence outcome. However, this gives rise to the question of confoundingselection with outcome.

The most common method for obtaining comparable but nonrandomized groups has been by matching. Matching tends to be done on only a few variables, since matching on multiple variables quickly tends to fail to find suitable matches, and, therefore, neither maximizes the available pool of patients for comparison, nor extensively matches them. Further, if one matches on variables known to influence outcome, these variables tend not to be able to be used as adjustments in the comparison. In this study the method of matching was according to propensity score. The strength of matching using the propensity score is that it simultaneously matches on all known characteristics, allows all variables to be used as risk factors, and maximizes the use of patients. However, sometimes one finds subsets of patients that have either an extremely low or high propensity for one or the other arms of the comparison and cannot be used. Thus, in this study matching according to propensity score eliminated 3976 of the single ITA patients from the propensity-matched analyses because their propensity scores predicted they were poorly comparable with the bilateral ITA patients. Recently, the propensity score has been incorporated into multivariable comparison analyses as a variable, as in our study, although its properties are debated. ²⁷ This creates a hybrid model in which most of the values for variables in the model are from each individual patient, but also each patient has a variable whose score is determined by the general behavior of the overall group, namely, the propensity score. Importantly, this strategy does not produce a model that can be readily used for outcome prediction, as is true of traditional risk factor models. The limitation of all matching schemes, including propensity scores, is that the matching is done for known sources of selection bias, but cannot protect against unknown sources of selection basis, as do randomized trials.

In the hope of analyzing outcomes in ways that are both accurate and understandable, we have used multiple statistical strategies and have presented the results in multiple ways. Unadjusted late survival and reoperation-free survival curves for the entire patient sample are contained in Fig. 1, and the unadjusted survival curves and reoperation hazard curves for the propensity-matched patients are in Fig. 2. Unadjusted curves for the propensity-matched patients in regard to the risk factors of age, diabetes, and left ventricular function are contained in Figs. 5, 6, and 7. We have generated traditional parsimonious risk factor models for the entire patient group (Table II), parsimonious models for propensity-matched patients (Tables V, VI, and VI), and nonparsimonious models maximally incorporating all available variables as well as their propensity scores (Table III). Each type of model has advantages and disadvantages. The use of parsimonious risk factor models enables the identification of multiple risk factors that affect outcome and allows the production of models that can be used for outcome prediction. For the overall comparison of the surgical strategy of 1 versus 2 ITA grafts, nonparsimonious models offer the most conservative and reliable estimates because all available variables including the propensity score are included. The bootstrap resampling adjustment of the estimates is an attempt to make the comparisons even more conservative.

Every comparison has shown better outcomes for patients with bilateral ITA grafts. Multiple statistical techniques have been used in efforts to reject the idea that bilateral ITA grafting produces superior outcomes to single ITA grafting, but all have failed. They all consistently point to a benefit of bilateral ITA grafting, and the more adjustments that are made for confounding, the more the increase in the estimated magnitude of the benefit of receiving 2 ITA grafts. The answer to the simple question, "Are 2 ITA grafts better than one?" is "Yes."

Substantial effort was made to quantify the extent of benefit of bilateral ITA grafting. Fig. 4 illustrates the predicted distribution of outcome improvement over the first 12 postoperative years based on 1 versus 2 ITA grafts using multivariate techniques that included all patient information for the propensity-matched groups. Over that period, bilateral ITA grafting prevented reoperation more than it prevented death. For half of the patient sample, the improvement in the survival based on bilateral ITA grafting was predicted to be less than 6% (0.5% per year of follow-up).

For the practicing surgeon, identification of subgroups predicted to experience only small benefit from a more complex surgical technique is obviously useful. Traditionally important risk factors for late survival (age, diabetes, and left ventricular function) were examined to see whether these variables could be used to identify patients without major benefit. Bilateral ITA grafting achieved better survival regardless of left ventricular function and regardless of the presence or absence of treated diabetes. Young patients benefitted relatively little in terms of survival (over the first 12 postoperative years) but benefitted greatly from bilateral ITA grafting in the avoidance of reoperation and PTCA. No subgroups were identified that were predicted to have worse outcomes with bilateral ITA grafting.

Bilateral ITA grafting improves outcomes for patients at a high risk of an event. Patients at a high risk to die have better survival and patients at a high risk of reoperation and/or PTCA had fewer of these reinterventions. However, the variables that predict the composition of high and low benefit groups are multiple, as depicted in Figs. 8 and 9.

It is also important to re-emphasize that this study basically deals with a 12-year postoperative follow-up interval. Some patients, particularly in the younger age group, do not have a substantial increase in their already favorable survival over 12 years, but there is no sign that the benefits of bilateral ITA grafting are lessening with time, and by 20 postoperative years the improvement in survival may be much greater.

We examined the details of surgical strategy (Appendix 2) in both univariate and multivariate settings. Any ITA-LAD graft decreased the risk of death, reoperation, and PTCA, and there was a slight but significant increase in the risk of reoperations for patients with an ITA–RCA graft. Other studies have also noted imperfect outcomes with RITA–RCA grafts. ^28-30No other variations in the use of ITA grafts influenced outcomes. Examination of Appendix 2 will show some unusual grafting combinations that were usually based on unusual situations of coronary anatomy, myocardial infarction in the distribution of one of the major vessels, or ITA size or anatomy, but also may have been related to relative surgical inexperience. Our current strategy is to use the ITA grafts to revascularize the most important left-sided vessels, sometimes as a composite LITA and RITA "Y" graft. This is a change in strategy relative to this series and is based in part on the emergence of other arterial grafts (right gastroepiploic artery and radial artery) that may be used as grafts to the RCA. Composite ITA grafting allows more options in the use of the free RITA but was not used during the years encompassed in the study. Thus, although this series may not reflect the ideal use of the ITAs in every case, it does reflect the results of what was the actual practice for a consecutive series of patients. It is also important to note that this is not a study of total arterial revascularization—the majority of patients received some vein grafts, a factor that may or may not have compromised outcomes. Today we are more adept at achieving total arterial revascularization, but whether that will improve outcomes is not known.

This study has produced different results than previous studies that have compared single and bilateral ITA grafting. The simplest possible explanation for this difference is that this study contains larger numbers of patients who received bilateral ITA grafts and they have been followed up for longer postoperative intervals. A study from Europe by Sergeant, Blackstone, and Meyns ⁵ that investigated similar issues and contained large patient numbers showed no advantage for bilateral ITA grafting in regard to either survival or reoperation-free survival. However, although that study contained a large number of patients overall, the subset of patients who had undergone bilateral ITA grafting was much smaller and their mean postoperative follow-up interval was only approximately 6 years. Also, differences between the European and American medical milieus in the indications for primary bypass surgery and in the indications for reinterventions may influence outcomes. Smaller studies from United States centers have not shown significant survival differences, but many have shown either significant decreases in late cardiac events or trends in that direction associated with bilateral ITA grafting. ^6,8,12,13

We do not believe that the implication of these data is that every patient undergoing bypass surgery must receive bilateral ITA grafts. Emergency operations and reoperations were excluded from this study, and few patients over 70 years of age were included. There was an increased risk of wound complications for diabetic patients, although that increase in risk was small. Bilateral ITA grafting makes coronary surgery more complicated, and the results of this study must be evaluated with the realization that the surgeons involved were experienced in coronary surgery in general and in complex arterial grafting in particular. Concepts of coronary surgery continue to evolve, and the advent of radial artery grafts provides another choice for arterial revascularization, although the long-term effectiveness of radial artery grafts remains to be determined. Opponents of multiple arterial grafting strategies may argue that there were patients in our study for whom the incremental benefit of bilateral ITA grafting (at least over 12 years) may be small, particularly in terms of survival. Ultimately, for each bypass operation each individual surgeon must decide what operation that surgeon can accomplish for that patient at that time that will produce the best short- and long-term outcomes. Many factors must be considered.

However, until now it has been the position of some coronary artery surgeons that the consideration of multiple arterial grafting could be ignored because no clear evidence existed that outcomes were improved for any patient subsets. That position is no longer tenable. The data from this study indicate that the strategy of bilateral ITA grafting decreases the risk of death, reoperation,and PTCA when compared to a strategy of single ITA grafting.

	Appendix: Discussion

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 
Dr Hendrick B. Barner (St Louis, Mo). It is a pleasure to discuss this report, which I have long anticipated and which reflects surgical expertise and dedicated patient follow-up. It has been 12 years since the report of enhanced survival and reduced ischemic events when a single ITA is grafted to the LAD coronary artery. For a number of reasons, the logical step of showing an advantage for 2 ITAs over 1 ITA has been more arduous, but the dedication of the Cleveland Clinic surgeons has resulted in another landmark paper.

In most hearts, the LAD is the most important coronary artery, and it is self-evident that if 2 ITAs are to be proven better than 1 ITA, the second ITA must be placed to the next most important coronary artery in terms of the mass of viable myocardium subtended by that artery. By and large, the authors have achieved this goal, but not perfectly, since 7% of the single ITA grafts were placed to the diagonal artery or its equivalent, and in 13% of patients with bilateral ITA grafting, 1 ITA was placed to the diagonal artery or its equivalent. Thus 6% more of bilateral ITA patients had placement of 1 ITA to the diagonal artery, which was probably not the second most important coronary artery. Had it been placed to a more important coronary artery, this might have resulted in even better outcome for the bilateral ITA group.

The modest use of the ITA as a free graft is surprising considering the Cleveland Clinic was one of the first to report such use in 1973. As a consequence, the in situ RITA crossed the midline to reach the left side of the heart in 64% of patients. I believe your practice has been to bring it across anteriorly rather than through the transverse sinus, and you have reported relative success when some of these patients did require reoperation from the standpoint of injury to the anterior crossing RITA. In view of the favorable experience with routine use of the free RITA, as reported by Dr Tatoulis, has your practice in this regard changed in the last decade?

The incidence of diabetes was relatively high at 10.8% for the bilateral group and low at 13% for single ITA use. It is understandable that this incidence would be low in the bilateral group because of the risk of mediastinitis, but was it low in the single ITA group because of patient selection? This raises the issue of a selection by the surgeon, which is probably the greatest limitation of this study.

Dr Lytle. Let me address the statistical methods first. Two different statisticians used two different approaches to theses analyses. Ms Houghtaling used Cox proportional hazard models of the entire group and selected subgroups, whereas Dr Blackstone used hazard function and interaction based analyses. They ended up with almost identical conclusions. I am very comfortable with the statistical methods used in this study.

In regard to the vessels grafted with the ITA grafts: Although the LAD coronary artery is usually the most important anterior vessel, that is not always the case. In the single ITA group, 7% of patients received an ITA-diagonal graft, usually because the diagonal was the most important artery. These patients may have had a small LAD or a previous LAD infarct. For the bilateral ITA graft group, in 60% of patients the LAD and circumflex were the vessels receiving ITA grafts and in 20% it was the LAD and RCA. The hodgepodge of other combinations was usually reflective of unusual anatomic situations, a lack of LAD stenosis, or previous myocardial infarction. Remember, this was a consecutive series.

I share your concerns about patient selection and, indeed, the tendency to select good-risk patients for bilateral ITA grafting has always made it difficult to clearly demonstrate that surgical strategy produces better long-term outcomes. There is no question that there is a bias to select good-risk patients for extensive arterial grafting strategies. However, in this study the baseline differences in the single and bilateral ITA groups in the prevalence of variables like diabetes, age, and female gender were relatively small. We attempted to adjust for those baseline differences with our statistical analyses. It is impossible to ever rule out completely the fact that the improved survival of the bilateral ITA group was not based on some good-risk patient-related characteristics that we did not identify and adjust for. However, although the risk of late death is less for good-risk patients, previous studies have clearly shown that the prevalence of late reoperation is greater for good-risk patients. The patients in the bilateral ITA group had both fewer late deaths and fewer late reoperations. Thus we cannot have it both ways. If we say that the patients in the bilateral ITA group were better risk patients, it would stand to reason that they should have had more reoperations rather than fewer reoperations. In my opinion, these data are suggestive that survival is improved with the bilateral ITA strategy compared with a single ITA graft, but I think that there is no question that the risk of reoperation is a lot less when 2 ITAs are used.

	Acknowledgments

	References

Top Abstract Introduction Patients and methods Results Discussion Appendix: Discussion References

 

1. Lytle BW, Loop FD, Cosgrove DM, Ratliff NB, Easley K, Taylor PC. Long-term (5-12 years) serial studies of internal mammary artery and saphenous vein coronary bypass grafts. J Thorac Cardiovasc Surg 1985;89:248-58.
   
2. Loop FD, Lytle BW, Cosgrove DM, et al. Influence of the internal-mammary-artery graft on 10-year survival and other cardiac events. N Engl J Med 1986;314:1-6.
   
3. Cameron A, Davis KB, Green GE. Coronary bypass surgery with internal-thoracic-artery grafts. Effects on survival over a 15-year period. N Engl J Med 1996;334:216-9.
   
4. Okies JE, Page US, Bigelow JC, et al. The left internal mammary artery: the graft of choice. Circulation 1984;70(Suppl):I213-21.
   
5. Sergeant P, Blackstone E, Meyns B. Validation and interdependence with patient-variables of the influence of procedural variables on early and late survival after CABG. K. U. Leaven Coronary Surgery Program. Eur J Cardiothorac Surg 1997;12:1-19.
   
6. Morris JJ, Smith LR, Glover DD, et al. Clinical evaluation of single versus multiple mammary artery bypass. Circulation 1990;82 (Suppl):V214-23.
   
7. Accola KD, Jones EL, Craver JM, et al. Bilateral mammary artery grafting: avoidance of complications with extended use. Ann Thorac Surg 1993;56:872-9.
   
8. Pick AW, Orszulak TA, Anderson BJ, Schaff HV. Single versus bilateral internal mammary artery grafts: 10-year outcome analysis. Ann Thorac Surg 1997;64:599-605.
   
9. Lytle BW, Cosgrove DM, Loop FD, Borsh J, Goormastic M, Taylor PC. Perioperative risk of bilateral internal mammary artery grafting: analysis of 500 cases from 1971 to 1984. Circulation 1986;74(Suppl):III37-41.
   
10. Cosgrove DM, Lytle BW, Loop FD, et al. Does bilateral internal mammary artery grafting increase surgical risk? J Thorac Cardiovasc Surg 1988;95:850-6.
    
11. Lytle BW, Cosgrove DM, Saltus GL, Taylor PC, Loop FD. Multivessel coronary revascularization without saphenous vein: long-term results of bilateral internal mammary artery grafting. Ann Thorac Surg 1983;36:540-7.
    
12. Fiore AC, Naunheim KS, Dean P, et al. 1990: Results of internal thoracic artery grafting over 15 years: single versus double grafts. Ann Thorac Surg 1990;49:202-9.
    
13. Naunheim KS, Barner HB, Fiore AC. Results of internal thoracic artery grafting over 15 years: single versus double grafts. 1991 update. Ann Thorac Surg 1992;53:716-8.
    
14. Berreklouw E, Schonberger JP, Ercan H, et al. Does it make sense to use two internal thoracic arteries? Ann Thorac Surg 1995;59:1456-63.
    
15. Galbut DL, Traad EA, Dorman MJ, et al. Twelve-year experience with bilateral internal mammary artery grafts. Ann Thorac Surg 1985;40:264-70.
    
16. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958;53:457-81.
    
17. Blackstone EH, Naftel DC, Turner ME Jr. The decomposition of time-varying hazard into phases, each incorporating a separate stream of concomitant information. J Am Stat Assoc 1986;81:615-24.
    
18. Rosenbaum P, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41-55
    
19. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15:361-87.
    
20. Efron B, Tibshirani R. Bootstrap methods for standard errors, confidence intervals, and other measures of statistical accuracy. Stat Sci 1986;1:54-77.
    
21. Rubin DB. Estimating causal effects from large data sets using propensity scores. Ann Intern Med 1997;127:757-63.
    
22. Cook EF, Goldman L. Performance of tests of significance based on stratification by a multivariate confounding score or by a propensity score. J Clin Epidemiol 1989;42:312-24.
    
23. Cox DR Regression models and life tables. J R Stat Soc 1972;34:187-202.
    
24. Tector AJ, Kress DC, Downing FX, et al. Complete revascularization with internal thoracic artery grafts. Semin Thorac Cardiovasc Surg 1996;8:29-41.
    
25. Dion R, Etienne PY, Verhelst R, et al. Bilateral mammary grafting: clinical, functional and angiographic assessment in 400 consecutive patients. Eur J Cardiothorac Surg 1993;7:287-93.
    
26. Carrel T, Horber P, Turina M, et al. Operation for two-vessel coronary artery disease: midterm results of bilateral ITA grafting versus unilateral ITA and saphenous vein grafting. Ann Thorac Surg 1996;62:1289-94.
    
27. Drake C, Fisher L. Prognostic models and the propensity score. Int J Epidemiol 1995;24:183-7.
    
28. Dietl CA, Benoit CH, Gilbert CL, et al. Which is the graft of choice for the right coronary and posterior descending arteries? Comparison of the right internal mammary artery and the right gastroepiploic artery. Circulation 1995;92(Suppl):II92-7.
    
29. Chow MS, Sim E, Orszulak TA, et al. Patency of internal thoracic artery grafts: comparison of right versus left and importance of vessel grafted. Circulation 1994;90(Suppl):II129-32.
    
30. Lytle BW, Cosgrove DM. Coronary artery bypass surgery. In: Wells SA, editor. Current problems in surgery. Volume 29. St Louis: Mosby-Year Book; 1992. p. 733-807.
    

This article has been cited by other articles:

				B. Marcheix, F. Vanden Eynden, P. Demers, D. Bouchard, and R. Cartier Influence of diabetes mellitus on long-term survival in systematic off-pump coronary artery bypass surgery. Ann. Thorac. Surg., October 1, 2008; 86(4): 1181 - 1188. [Abstract] [Full Text] [PDF]

				J. F. Sabik III, A. Stockins, E. R. Nowicki, E. H. Blackstone, P. L. Houghtaling, B. W. Lytle, and F. D. Loop Does Location of the Second Internal Thoracic Artery Graft Influence Outcome of Coronary Artery Bypass Grafting? Circulation, September 30, 2008; 118(14_suppl_1): S210 - S215. [Abstract] [Full Text] [PDF]

D. Glineur, C. Hanet, A. Poncelet, W. D'hoore, J.-C. Funken, J. Rubay, J. Kefer, P. Astarci, V. Lacroix, R. Verhelst, et al. Comparison of Bilateral Internal Thoracic Artery Rev