## Journal of the American College of Cardiology

# Early mortality and morbidity of bilateral versus single internal thoracic artery revascularization: propensity and risk modeling

## Author + information

- Received April 27, 2000.
- Revision received August 24, 2000.
- Accepted October 4, 2000.
- Published online February 1, 2001.

## Author Information

- John P.A Ioannidis, MDâ,Â§,
- Othon Galanos, MDâ ,Â§,
- Demosthenes Katritsis, MD, PhD, FACCâ¥,
- Cliff P Connery, MDâ ,
- George E Drossos, MDâ¡,
- Daniel G Swistel, MDâ and
- Constantine E Anagnostopoulos, MDâ ,* (cea8{at}columbia.edu)

- âµ*Reprint requests and correspondence: Dr. Constantine E. Anagnostopoulos, Professor of Surgery and Senior Attending, Cardiothoracic Surgery, St. LukeâsâRoosevelt Hospital Center, Columbia University College of Physicians and Surgeons, 1111 Amsterdam Ave., New York, New York 10025

## Abstract

OBJECTIVES

We examined whether bilateral internal thoracic artery (BITA) revascularization is associated with any increased in-hospital mortality and complications compared with single internal thoracic artery (SITA) revascularization.

BACKGROUND

Despite proven long-term benefits, BITA revascularization has been slow to be adopted because of fear of increased early morbidity.

METHODS

We evaluated 1,697 consecutive patients undergoing BITA (n = 867) or SITA (n = 830) revascularization. We used propensity score analyses and adjusted risk models to address differences between arms.

RESULTS

There were 20 (2.3%) deaths in the BITA group versus 26 (3.1%) in the SITA group (odds ratio 0.73, p = 0.30). Propensity analysis identified several parameters that affected the decision to use BITA. Adjusting for propensity score and all potential risk factors, the odds ratio for death with BITA versus SITA was practically 1. Bilateral internal thoracic artery revascularization did not increase the number of in-hospital complications with the possible exception of deep sternal wound infections (11 [1.3%] vs. 3 [0.4%], p = 0.057). In multivariate modeling BITA increased the risk of deep sternal wound infections only in emergent cases and in older patients; the excess risk was negligible among 1,206 patients (71.1% of total) who did not have emergent revascularization and were ≤70 years old (risk difference 0.3%, p = 0.74). There was no difference in length of stay after adjustment for propensity factors (mean 11.3 vs. 11.7 days, p = 0.66).

CONCLUSIONS

Bilateral internal thoracic artery revascularization grafting confers no increased risk for early death and does not prolong hospital stay. The small increase in the risk of deep sternal wound infections does not affect the majority of patients.

Recently evidence has accumulated to show that bilateral internal thoracic artery (BITA) revascularization offers significant long-term survival benefits as compared to single internal thoracic artery (SITA) revascularization in patients receiving coronary artery bypass grafts (CABG) (1–7). Nevertheless, BITA revascularization has not been universally adopted, and many surgeons and referring cardiologists are reluctant to make it a routine procedure. One of the potential concerns is that BITA may be associated with excess in-hospital mortality and morbidity because it may require more time and skill to perform (8). However, such excess morbidity and mortality have not been documented beyond speculation. Furthermore, adoption of the technique may have varied among different surgeons who feel more or less comfortable with using BITA grafts. Finally, in the absence of randomized trials, the selection of the patients who undergo BITA rather than SITA is often guided by the knowledge of specific preoperative factors that may be perceived to be predictors of poor outcome (1).

In order to obviate the concerns about BITA revascularization, one would need to examine a large number of patients undergoing CABG with BITA or SITA and adjust for selection biases, surgeon preferences and risk factors. We undertook this task using a large database encompassing the four-year experience of a busy academic cardiothoracic center in New York City. Our main aim was to estimate as precisely as possible whether there is any excess early mortality and morbidity with BITA versus SITA revascularization and, if so, what the magnitude of this excess is and how it would compare to the known long-term benefits of BITA revascularization.

## Patients and methods

### Patient population

We analyzed 1,697 consecutive patients who were discharged or died between January 1, 1993, and December 31, 1996, after undergoing primary isolated revascularization either with BITA or with SITA plus vein graft(s) at St. Luke’s–Roosevelt Hospital Center, a university hospital of the Columbia University College of Physicians and Surgeons. This number excluded cases where a SITA was used as a single graft. There were 830 qualifying SITA and 867 qualifying BITA revascularization procedures during this period. The data were prospectively entered in the New York State Adult Cardiac Surgery Reports and have been carefully double-checked for accuracy. The years included contain the maximum consecutive and easily retrievable data where BITA (not “double artery”) was listed as separate entry code.

### Predictors of procedure selection and risk

For each patient, information was collected on several potential predictors of procedure selection and risk corresponding to the New York State Adult Cardiac Surgery Reports (Table 1). Information was complete on all parameters shown in Table 1, except for ejection fraction missing on 37 patients. On the basis of the profile of these patients, we imputed a median value of 40% in these cases, slightly below the cohort median (45%).

### Outcomes

For each patient, information was available on the main outcome of interest, namely death during the hospitalization (including death after transfer to another facility) and on several complications. These included stroke, transmural myocardial infarction (MI), nontransmural MI, deep sternal wound infection, bleeding requiring reoperation, return to the operating room for second cardiopulmonary bypass, heart block requiring permanent pacemaker, sepsis or endocarditis, gastrointestinal bleeding, perforation or infarction, renal failure requiring dialysis, intraaortic balloon pump (IABP) insertion during or after surgery, left ventricular assist device (LVAD) insertion during or after surgery, respiratory failure, brachial plexus impairment, phrenic nerve palsy and malignant ventricular arrhythmia. Information was also collected on the length of stay from day of surgery to discharge.

### Propensity analysis

Given the nonrandomized character of the study, it was anticipated that several parameters might have influenced the decision to use BITA rather than SITA. These parameters may have included several postulated or perceived predictors of risk. In addition, the choice of BITA versus SITA would also be dependent on the specific surgeon performing the procedure. In order to model and address the resulting selection biases, we performed multivariate propensity modeling with logistic regressions considering all the potential variables that may affect the choice of procedure (9,10). The modeled outcome was the selection of BITA grafting. The multivariate model initially considered all the known potential predictors of risk of death/complications described in the previous section, the calendar time (to allow for more widespread use of BITA over time), and separate covariates for individual surgeons, and it was reduced to a more parsimonious model by backward elimination of variables. This parsimonious multivariate model also estimated a propensity score for each patient. The propensity score is the likelihood that a patient would undergo BITA grafting given the profile of his/her risk factors and the surgeon undertaking the procedure. The propensity score can range from 0% to 100%. Patients were sorted into quintiles of propensity and we examined within each quintile whether both BITA and SITA were adequately represented (>10% each). Propensity-score matching (9,10)allows to comparatively examine BITA versus SITA patients with a similar profile as to the characteristics that are likely to influence the decision to select BITA. As suggested (9,10), we also performed analyses limited to quintiles where both BITA and SITA were adequately represented to allow a more meaningful comparison between procedures. The propensity score was also used as an adjusting parameter in some of the analyses of the risk factor models.

### Risk factor analysis for mortality and morbidity

Differences in risk factors for hospital death and deep sternal wound infection in the BITA versus the SITA groups may affect the comparative outcome of patients undergoing each of the two procedures if they act as confounders. Therefore, for each outcome of interest, we performed multivariate analyses adjusting for various parameters. A number of different models (11)were considered: 1) a model adjusting only for the propensity score mentioned above; 2) a model adjusting for all potential risk factors as well as for separate covariates for individual surgeons; 3) same, but without covariates relating to individual surgeons; 4) same as model 2, but also adjusting for the propensity score; 5) same as model 3, but also adjusting for the propensity score; and 6) a backward elimination model considering initially all the variables of model 3 and proceeding to a more parsimonious form by dropping variables according to likelihood ratio criteria. Starting from the latter model we added interaction terms between age and selected risk factors and between BITA and selected risk factors and again performed backward elimination of variables. BITA was forced-entered in the backward elimination models whenever it was not retained either alone or as an interaction term. The final derived models with interaction terms were used for estimating the probability of death and the probability of deep sternal wound infection for each patient.

Multivariate logistic models were estimated based on maximization of likelihood (12). Continuous variables were examined for linearity of risk functions. For the predicted quartiles of risk for the final death risk model, we estimated the extreme quartile odds ratio (the odds ratio for death in the upper quartile of predicted risk vs. in the lower quartile of predicted risk). This metric is useful in assessing the heterogeneity of risk in the patient population (13). BITA was compared with SITA within each quartile of predicted risk of death.

### Length of stay

The length of stay was defined as the number of days from the day of the surgery until hospital discharge (or death). For the comparison of BITA versus SITA, we performed both an unadjusted analysis by Student *t*test and adjusted analyses using a generalized linear model approach for analysis of variance. All parameters that were statistically significant in the propensity analysis were initially considered as covariates so as to account for all variables that may have influenced the decision to perform BITA rather than SITA. At a second stage, we considered only variables with p < 0.10 in the initial model. A linear regression analysis was also performed using the length of stay as dependent variable and BITA versus SITA as well as the selected covariates as independent variables.

All analyses were conduced in SPSS version 9.0 (SPSS Inc, Chicago, Illinois) and in StatXact 3 (Cytel Inc, Boston, Massachusetts). Exact p values are provided when numbers are small. All p values are two-tailed.

## Results

### Direct comparison of BITA versus SITA groups

Table 1shows the comparison of the BITA versus SITA groups for selected parameters. As shown, the two groups were markedly different in regards to several potential risk factors, which may have influenced the decision of the surgeons. Table 2shows a direct comparison of the two groups for the main outcomes of interest and for other secondary outcomes. In these unadjusted analyses, there were no statistically significant differences for death or for any other complications, with the exception of significantly fewer IABP inserted during or after surgery in the BITA group, probably a reflection of selection bias. There were also some notable trends. There were more patients without any complications at all in the BITA group (86.0% vs. 82.5%), fewer nontransmural MIs in the BITA group (2.4% vs. 4.1%), and fewer deep sternal wound infections in the SITA group (0.4% vs. 1.3%). There was no significant difference in hospital mortality (2.3% vs. 3.1%). These comparisons are unadjusted for several characteristics that may have affected the selection of BITA versus SITA for the revascularization procedure and should therefore be interpreted with caution.

### Propensity analysis

Table 3gives the results of the multivariate analysis of predictors of choosing BITA rather than SITA revascularization. The chances of BITA being performed instead of SITA increased when the patient was younger, male, a smoker in the past year, had a high ejection fraction and when the left circumflex artery was diseased. On the other hand, a number of preoperative risk factors seemed to restrain surgeons from performing BITA grafting. These included a history of stroke, carotid/cerebrovascular disease, diabetes, immunodeficiency, renal failure with elevated creatinine, an emergent situation and most of all emergency transfer to the operating room after PTCA. Calendar time was also an important predictor, as the use of BITA increased over time. Patients with previous open-heart surgery were also less likely to receive BITA; to some extent this was due to the fact that BITA might not be a feasible option if one internal thoracic artery had already been used.

There was <10% representation of the less frequent revascularization procedure group in the extreme propensity quintiles estimated by the multivariate propensity model. In the low-propensity quintile 26 patients (7.7%) underwent BITA revascularization, whereas among the 339 patients in the high-propensity quintile there were 23 patients (6.8%) with SITA revascularization. Excluding the extreme (high and low) propensity quintiles, where matching would be difficult due to poor representation of one group, there was no statistically significant differences for the remaining 525 patients undergoing BITA versus the 494 patients undergoing SITA grafting either for mortality (8 vs. 14) or for deep sternal wound infections (7 vs. 2). The odds ratio (OR) for both outcomes (0.53 [95% CI, 0.22–1.28, p = 0.20] and 3.32 [95% CI, 0.69–16.1, p = 0.18]) were reassuringly similar to those obtained from all 1,697 patients.

## Risk factor analysis

### Hospital mortality

As shown in Table 4, there was no statistically significant difference between the BITA and SITA groups regardless of the parameters used for adjustment. In the three propensity quintiles where there was adequate representation of both the BITA and SITA groups, there was no evidence of excess mortality in the BITA group.

The final model with interaction terms showed that besides a history of previous open-heart operation(s) (OR 2.84, p = 0.021), age was also a very strong independent predictor of mortality (OR 1.055 per year, p = 0.006) and it interacted with a history of femoral/popliteal vascular disease (OR 1.013 per year, p = 0.017), diabetes (OR 1.009 per year, p = 0.050), high creatinine (OR 1.031 per year, p < 0.001), left main disease (OR 1.011 per year, p = 0.025) and emergent operation (OR 1.018 per year, p = 0.014) in increasing the risk of death, while it interacted with ejection fraction in decreasing the predicted risk of death (OR 0.9995 per 1 point increase in the [ejection fraction × age] product, p = 0.012). BITA did not significantly increase the risk of death in this model either independently (OR 1.14, p = 0.69) or in interaction terms.

Table 5shows separate analyses for the comparison of BITA versus SITA in terms of hospital mortality in each quartile of predicted risk based on this final model. The extreme quartile OR for death was 8.4, suggesting the definition of a moderately heterogeneous population for the study cohort. Actual event rates were similar in the three quartiles of low to moderate predicted risk. There was no strong evidence that BITA carried a clinically meaningful higher risk of hospital mortality in any of the four quartiles. Thus there was no strong risk by BITA interaction.

### Deep sternal wound infections

As shown in Table 4, deep sternal wound infections were significantly more common in the BITA group regardless of the method of adjustment in the analysis. The final backward elimination model, considering interaction terms, showed that the risk of deep sternal wound infection increased among patients with emergent operation who had BITA (OR 7.28, p = 0.015), among more elderly patients who had BITA (OR 1.023 per year, p = 0.024) and among elderly patients with diabetes (OR 1.022 per year, p = 0.010). The final logistic model was: logit (deep sternal wound infection) = 1.99 (BITA × emergent operation) + 0.0297 (age × BITA) + 0.0214 (age × diabetes) − 6.51. Taking all these factors into account, the risk of deep sternal infection was clinically and statistically significantly increased with BITA only in elderly patients and those with emergent operation. Among the 1,206 patients without emergent operation and age up to 70 years, the rates of deep sternal wound infection were 6/667 versus 3/539 in the BITA versus SITA groups (0.9% vs. 0.6%, risk difference 0.3%, 95% CI, −0.6% to 1.3%). There was no difference even among diabetics within this patient group (1.2% [2/161] vs. 0.8% [2/238]). On the other hand, among the 491 patients of age 71 or older and/or emergent operation, the rates were 5/200 with BITA versus 0/291 with SITA with a 2.5% excess (95% CI, 0.3 to 4.7%) in the BITA group; however there was no difference in mortality (5.0% vs. 4.5%, p = 0.83) among these 491 patients. Moreover, no deep sternal wound infections occurred among 26 BITA patients aged 80 or older. No other parameters were significantly related with deep sternal wound infections.

## Length of stay

The overall mean length of stay during these “pre-managed care” years (1993–1996) was 10.33 (±0.42) days in the BITA group versus 12.66 (±0.55) days in the SITA group. This corresponded to 2.33 fewer days of stay in the BITA group (95% CI, 0.98–3.68), which was statistically significant in the unadjusted analysis (p = 0.001). The difference did not reach formal statistical significance in any quintile of propensity score (mean length of stay values for the BITA vs. SITA groups: 18.5. vs. 15.5 [p = 0.49], 11.0 vs. 11.2 [p = 0.89], 9.4 vs. 10.8 [p = 0.12], 10.8 vs. 11.0 [p = 0.90], 9.5 vs. 8.5 [p = 0.69], in the five quintiles of lower to higher BITA propensity).

After adjusting for all variables that were significantly related to the propensity of performing BITA rather than SITA, there was no significant difference in the length of stay between the SITA and BITA groups. At mean covariate values, the adjusted length of stay was 11.2 (±0.52) days in the BITA group (p = 0.62) versus 11.6 (±0.52) days in the SITA group. Statistically significant predictors of the length of stay in this analysis of variance included age (p < 0.001), ejection fraction (p = 0.024), renal disease with creatinine >2.5 mg/dl (p < 0.001), and emergent operation (p = 0.011). The regression model considering these four parameters as well as BITA/SITA showed that the length of stay increased by 0.21 days per 1 year of increasing age, 4.2 days in the presence of emergent operation and 11.5 days in the presence of an elevated creatinine, and decreased by 0.57 days per each 10% increase in the ejection fraction, whereas it was nonsignificantly shortened by 1.17 days in patients with BITA grafts.

## Discussion

### Advantages and concerns regarding BITA

Accumulating evidence has shown that BITA revascularization confers strong long-term benefits (1–7). The largest series of patients comparing BITA versus SITA grafts has shown that 5-, 10-, and 15-year mortality rates are significantly decreased by about one-third in the BITA patients (1). Similarly, SITA is disadvantageous when reoperation and percutaneous transluminal coronary angioplasty outcomes are considered in the long term. Similar findings have been presented by other investigators as well (2,3,5–7). Despite this evidence and the theoretical support to the superiority of BITA over SITA grafting (where vein grafts also have to be used), there are still reservations among cardiologists and cardiothoracic surgeons as to whether BITA should be adopted as the routine procedure of coronary revascularization. There may have been concerns that BITA may take longer to perform, may be dependent on surgical skill and may adversely affect early in-hospital mortality and morbidity, in particular deep sternal wound infections.

### Main findings

Our study counters these reservations. Using a large series of consecutive patients from our experience at a large cardiothoracic center, we have shown that BITA confers no increased risk of in-hospital mortality. Using a variety of different statistical methods to account for the potential effects of selection biases, individual surgeons and confounding, we always found that the probability of in-hospital death is practically identical in the BITA and SITA groups.

### Deep sternal wound infections are a concern only for a minority of patients

There was no evidence that BITA increases the risk of in-hospital complications. There was only one exception to this rule: deep sternal wound infections were more common in the BITA group. However, the absolute risk difference was very small and in multivariate modeling it seemed that the risk difference was nonexistent or negligible for the large majority of patients. Among 300 patients ≤70 years old and not requiring emergent operation, one might see a single excess case of deep sternal wound infection with BITA—and even this excess is uncertain and may have been due to chance. By contrast, one excess case of deep sternal wound infection would be seen among 40 patients older than 70 and/or with emergent operation. In the same multivariate model, diabetes increased the risk of deep sternal wound infection, but this was true for both BITA and SITA patients. Two other smaller studies have shown no excess risk of deep sternal wound infections with BITA versus SITA in diabetics (14,15). We found no evidence that BITA confers a disadvantage in terms of increased infection risk for any of the categories of patients defined by preoperative factors that weighted heavily in the minds of surgeons in avoiding a BITA procedure. Such factors included renal impairment, other vascular disease, and female gender. The risk of deep sternal wound infection should not be viewed as a reason to avoid BITA revascularization in diabetics or other patients at increased risk for such infections.

### Length of stay is not increased with BITA

Further support for the superiority of BITA comes from the analysis of the length of stay. BITA did not prolong hospital stay; if anything there was a suggestion that BITA may actually shorten it. Various analyses considering parameters that may reflect selection bias and confounding reached the same conclusion. Thus there is further evidence that even the immediate in-hospital cost of the procedure may be less than (or at least the same as) the cost of SITA revascularization.

### Integration of other evidence

One other large study (1)has examined multivariate-adjusted in-hospital mortality with BITA versus SITA revascularization. The in-hospital mortality rate was exceptionally low in that series and the analysis was limited to a multivariate adjusted model which showed no excess mortality with BITA (odds ratio 0.81 [95% CI, 0.45–1.46]). In-hospital morbidity was not analyzed. A meta-analysis of that study with the multivariate adjusted mortality results of our cohort gives a Mantel–Haenszel pooled odds ratio of 0.91 with 95% confidence intervals of 0.60–1.38. The confidence intervals exclude the possibility of any clinically meaningful harm from the use of BITA revascularization that would match the long-term benefits.

### Study limitations

Our study did not use a randomized design, but no study to date in this field has been randomized. The data were entered prospectively and rigorously double-checked for accuracy. Given the observational design, we took due measures to account for sources of imbalance in the two compared strategies. Analyses using propensity scores, multivariate risk adjustment and adjustment for surgeon selection biases all reached similar conclusions. Another possible limitation is that there was a limited number of deep sternal wound infections in our cohort; thus the coefficients of the derived models may need some further validation in other cohorts. However, the large picture is unlikely to change much. Exploratory analyses using cross-validation (building the risk models in a subset of the cohort and validating the models in the remaining patients) yielded fairly similar results. Furthermore, the low rates of infectious complications further strengthen the notion that the fear of deep sternal wound infections should not restrain surgeons from performing BITA revascularization. Given the strong evidence regarding the superiority of BITA in the long term and the current data suggesting that BITA adds no significant problems in the short term, we think that at this point a randomized trial would probably be unethical for the majority of patients, although it may be appropriate for selected subgroups of patients, for example the very old and those who undergo emergent revascularization.

### Future directions

Newer techniques eliminating sternotomy risks, such as minimally invasive direct coronary artery bypass (16,17), port access (18,19)or robotic surgery (20,21)should allow us to further minimize any difference deriving from sternal wound infections. However, even without these future benefits there is strong rationale for adopting BITA revascularization for the majority of patients undergoing CABG.

## Acknowledgements

We are grateful to Madeleine Vitale Svitak, BS, MS, and Konstantine G. Ioannides, PhD, for their help in the preparation of the databases and early analyses.

- Abbreviations
- BITA
- bilateral internal thoracic artery
- CABG
- coronary artery bypass graft
- CI
- confidence interval
- IABP
- intraaortic balloon pump
- MI
- myocardial infarction
- LVAD
- left ventricular assist device
- OR
- odds ratio
- PTCA
- percutaneous transluminal coronary angiography
- PVD
- peripheral vascular disease
- SITA
- single internal thoracic artery

- Received April 27, 2000.
- Revision received August 24, 2000.
- Accepted October 4, 2000.

- American College of Cardiology

## References

- ↵
- ↵
- Cameron A.A,
- Green G.E,
- Brogno D.A,
- Thornton J

- Green G.E,
- Swistel D.G,
- Cameron A.A

- ↵
- ↵
- Rosenbaum P,
- Rubin D.B

- ↵
- ↵
- Hosmer D.W,
- Lemeshow S

- ↵
- Ioannidis J.P,
- Lau J

- ↵
- Hirotani T,
- Kameda T,
- Kumamoto T,
- Shirota S,
- Yamano M

- ↵
- ↵
- ↵