- Email: [email protected]
Relationship of Body Mass Index With Outcomes After Coronary Artery Bypass Graft Surgery
CARDIOVASCULAR
Relationship of Body Mass Index With Outcomes After Coronary Artery Bypass Graft Surgery Brandie D. Wagner, MS, Gary K. Grunwald, PhD, John S. Rumsfeld, MD, PhD, James O. Hill, PhD, P. Michael Ho, MD, PhD, Holly R. Wyatt, MD, and A. Laurie W. Shroyer, PhD Division of Cardiac Research, Department of Veterans Affairs Medical Center, Eastern Colorado Health Care System; Departments of Preventative Medicine and Biometrics, and Medicine, and Center for Human Nutrition, University of Colorado at Denver and Health Sciences Center, Denver, Colorado
Background. A debate exists whether obesity is a risk factor for operative mortality after coronary artery bypass graft surgery (CABG). The contradictory findings in the literature may largely be attributable to the variety of methodological approaches used to model the association between body mass index (BMI) and post-CABG outcomes. This study aims to investigate this association, and to uncover possible explanations for the lack of consensus across prior studies. Methods. Data were prospectively collected on 80,792 patients who underwent a CABG procedure during a 14-year period at the 45 Department of Veterans Affairs cardiac surgery programs. Generalized additive models were used to estimate the relationship of BMI and outcomes after a CABG procedure. Results. We found that the relationship of BMI with post-CABG mortality and morbidity is U-shaped with the minimum risk located around a BMI of 30 kg/m2,
indicating that patients classified as overweight have the lowest risk, and those in the lower end of the obese range do not have seriously elevated risk. This U-shape relationship is significantly nonlinear and robust to adjustment for other risk factors. Conclusions. This study demonstrates that BMI is an independent predictor of mortality and morbidity after CABG surgery. Previous studies that model BMI linearly or as categories cannot accurately capture this U-shaped relationship and are unlikely to find a significant contribution by including BMI. Further research is needed to determine the mechanisms of risk for patients with low and high BMI and whether interventions to modify BMI may improve patient outcomes.
T
adverse outcomes and prolonged hospitalization after a CABG operation. Thus, it remains unclear whether obesity is an independent risk factor for post-CABG outcomes, and whether obesity is an important target for risk modification to improve outcomes after CABG surgery. A significant limitation of some previous studies examining the relationship between obesity and postCABG outcomes has been the failure to evaluate potential nonlinear associations between body mass index (BMI) and outcomes. In addition, there has been limited formal evaluation of interactions between BMI and key covariates such as age, smoking, and diabetes. Accordingly, the objective of this study was to evaluate, in a large multicenter cohort, the relationship between BMI before surgery with post-CABG outcomes using generalized additive models (GAM). These models provide a flexible method for modeling nonlinear covariate effects, including covariate adjustments and interactions. It is hoped that the results of this study will help clarify the relationship between obesity and post-CABG outcomes, and determine whether BMI is a potential target for interventions to improve patient outcomes for CABG surgery.
he well-documented epidemic of obesity in the United States has led to increased focus on obesity as a risk factor for adverse health outcomes [1–7]. Previous studies have shown that obesity is associated with the development of chronic health conditions including diabetes and coronary artery disease. Moreover, obesity leads to disability and impaired patient quality of life, and contributes to escalating health-care costs [8, 9]. Despite this, controversy remains about the relation between obesity and outcomes among patients with cardiovascular disease. In particular, there is ongoing debate about obesity as a risk factor for adverse outcomes after cardiovascular procedures [6, 10 –16]. Several prior studies have found that obesity is not a risk factor for operative mortality after coronary artery bypass graft (CABG) surgery [10 –12, 14, 15], and it has been suggested that any risk from obesity is directly attributable to clustered risk factors of smoking or diabetes. Other studies [13, 16] have concluded that the extreme obesity category was a significant independent predictor for Accepted for publication March 5, 2007. Address correspondence to Dr Shroyer, Cardiac Research, Denver Department of Veterans Affairs Medical Center, 820 Clermont St (112R), Denver, CO 80220; e-mail: [email protected].
© 2007 by The Society of Thoracic Surgeons Published by Elsevier Inc
(Ann Thorac Surg 2007;84:10 –16) © 2007 by The Society of Thoracic Surgeons
0003-4975/07/$32.00 doi:10.1016/j.athoracsur.2007.03.017
11
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Table 1. Patient Risk Characteristics Risk Variable BMI (mean, SD) Age (mean, SD) Male (%) Current smoker (%) Comorbidities Diabetes (%) Normal Diet-controlled Insulin-controlled Renal insufficiency (%) COPD (%) CVD (%) Cardiac history Prior MI (%) Prior heart surgery (%) Surgical priority (%) Heart failure (%) Angina (%) Outcome variables 30-day mortality (%) 30-day morbidity (%) Endocarditis (%) Renal failure (%) Mediastinitis (%) Bleeding (%) Ventilator (%) RCB (%) Coma (%) Stroke (%) Cardiac arrest (%)
Overall (n ⫽ 80,792)
BMI ⬍ 18.5 (n ⫽ 887)
BMI 18.5–25 (n ⫽ 18,130)
BMI 25–30 (n ⫽ 34,063)
BMI 30–35 (n ⫽ 19,391)
BMI ⬎ 35 (n ⫽ 8,321)
28.6 (5.1) 63.5 (9.3) 99.1 30.4
16.2 (2.1) 64.9 (9.0) 98.9 45.4
22.9 (1.6) 65.0 (9.4) 99.0 38.8
27.4 (1.4) 64.1 (9.2) 99.3 29.6
32.1 (1.4) 62.3 (9.0) 99.1 26.0
38.6 (3.5) 60.4 (8.8) 98.5 24.4
66.8 19.1 14.1 16.2 23.6 20.2
77.0 14.0 9.0 15.9 37.5 25.9
77.7 12.5 9.8 16.8 27.2 23.2
69.5 18.1 12.4 16.3 21.5 20.4
58.6 23.7 17.7 15.8 22.7 18.5
49.7 27.7 22.6 15.2 24.6 16.0
6.9 7.2 5.8 4.3 38.1
9.2 5.3 8.2 5.6 38.9
7.3 7.0 6.8 4.5 40.0
7.1 7.7 5.9 3.8 38.1
6.5 7.4 5.2 4.3 37.0
6.2 5.5 4.7 5.3 36.6
3.4 12.6 ⬍0.1 1.1 1.4 2.7 7.3 0.4 0.7 1.7 2.6
7.1 16.9 0 1.1 1.4 4.5 10.0 0.5 0.7 2.0 4.3
4.0 14.0 ⬍0.1 1.2 1.1 3.8 8.0 0.4 0.8 1.9 3.0
3.2 11.7 ⬍0.1 1.0 1.2 2.6 6.7 0.4 0.6 1.8 2.5
3.0 11.9 ⬍0.1 1.1 1.7 2.0 7.1 0.4 0.7 1.7 2.4
3.6 14.1 ⬍0.1 1.4 2.6 1.9 8.9 0.3 0.7 1.3 2.9
BMI ⫽ body mass index; CABG ⫽ coronary artery bypass graft; COPD ⫽ chronic obstructive pulmonary disease; disease; MI ⫽ myocardial infarction; RCB ⫽ repeat cardiopulmonary bypass; SD ⫽ standard deviation.
Material and Methods The Department of Veterans Affairs (VA) Continuous Improvement in Cardiac Surgery Program (CICSP) prospectively collects patient-level risk, procedural, and outcome data on all patients undergoing cardiac surgery at the 45 VA cardiac surgical centers. The goal of the CICSP is to provide performance reports to VA-based clinical care team members for use in self-assessment and selfimprovement quality of care initiatives. A description of this complete database and the variables captured has been published previously [17, 18]. For this analysis we selected records for all patients undergoing CABG-only procedures from October 1, 1991, to September 30, 2005. From these records, patients with a BMI greater than 58 were excluded owing to the small numbers and inaccurate height recorded for double amputees. No adjustment was made for those with low BMI. There were 67 patients with missing outcome variables and these were excluded, giving a total of 80,792 records. The CICSP project has received both institutional review board ap-
CVD ⫽ cerebrovascular
proval and Health Insurance Portability and Accountability Act waiver for analysis [Colorado Multiple Institutional Review Board protocol 98 – 611]. For this study 12 predictor variables were considered; refer to Table 1 for the list of these variables. All variables are collected within the CICSP and these were chosen a priori for this analysis on the basis of clinical relevance and reliability. Body mass index is a widely used index of obesity and is calculated as weight in kilograms divided by height in meters squared. A dichotomous variable for hypertension, overall smoking status (eg, never smoked versus smoked greater than 3 months or less than 3 months before surgery), and hypoalbuminemia [10] as measured by serum albumin less than 3.5 g/dL were not completely collected within CICSP until 2001 and were therefore included in a subanalysis. The primary outcome measures used in this analysis were 30-day operative mortality and 30-day morbidity (occurrence of at least one of the major complications listed in Table 1). Deaths are verified using the VA Beneficiary Indicator
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
12
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Records Locator System [19], which is a national death registry for all veterans. Because of the on-site data collection used by CICSP’s surgical clinical nurse coordinators, less than 1.2% of the records had any missing values of risk variables. These values were imputed using the median for continuous variables and the most frequent value for categorical variables, which is consistent with the CICSP approach [17, 18]. Hastie and Tibshirani [20] developed the GAM to extend the generalized linear model (including logistic regression) when the form of the relationship between covariates and the outcome is suspected to be nonlinear. Few assumptions are made about the structure of the association of BMI with the outcome, and the data are allowed to shape the functional form of the relationship. These models were fit using the ‘GAM’ function in the S-PLUS software version 6.1 (Insightful Corp, Seattle, WA). Generalized additive models were used in the logistic regression setting to estimate the nonlinear relationship of BMI with both 30-day mortality and morbidity separately. The optimal smoothing parameter for the analysis was chosen based on Bayes information criteria values [21]. As a result, a smoothing spline model with 3 degrees of freedom was used for all smooth model terms. In the preliminary analysis, univariate models containing only preoperative BMI were estimated. To assess the stability of the BMI curve to differences in other patient factors, two further types of analyses were done. First, multivariate logistic regression models (including BMI while adjusting for other risk variables) were estimated. Adjustment for other risk variables accounts for imbalances in risk factors across the range of BMI (eg, more smokers in the low BMI range or diabetics in the high BMI range). In these models, age and surgery year were included in nonlinear form. Second, models including interactions between BMI and other patient risk factors were estimated. These allow different nonlinear association patterns between BMI and the outcome for different risk groups (eg, smokers versus nonsmokers or diabetics versus nondiabetics). These models were fit by estimating a separate BMI curve for each level of the categorical risk variable. These analyses were then repeated using a more recent subset of the data (2001 to 2005), which included additional risk variables that were of interest for adjustment. These variables include hypertension, overall smoking status, and hypoalbuminemia. Using a semiparametric GAM, the estimated effect of BMI as a continuous risk factor is represented by a flexible function rather than a single parameter, so the best way to display the results is to plot the function. Pointwise standard error bands are used to show the precision of the estimated curve. To test the significance of the nonlinear effects, a likelihood ratio test was used. Significant results of this test indicate that the nonparametric smoothing spline fits the data significantly better than a linear function.
Ann Thorac Surg 2007;84:10 –16
Results The distributions across BMI classes of these risk factors for the patients included in this analysis are displayed in Table 1. Of the 80,792 patients, 887 (1.1%) were classified as underweight (BMI ⬍ 18.5), 18,130 (22.4%) were normal weight (18.5 ⱕ BMI ⬍ 25), 34,063 (42.2%) were overweight (25 ⱕ BMI ⬍ 30), 19,391 (24.0%) were obese (30 ⱕ BMI ⬍ 35), and 8,321 (10.3%) were morbidly obese (BMI ⱖ 35). Increasing BMI was associated with younger age, lower rates of smoking and chronic obstructive pulmonary disease (chronic obstructive pulmonary disease does increase slightly in the higher BMI categories), lower rates of previous myocardial infarction and cerebral vascular disease, lower rates of emergent surgical priority, and higher rates of insulin-controlled diabetes. In particular, for patients with BMI less than 18.5 kg/m2, 45% were current smokers and 38% had chronic obstructive pulmonary disease compared with 24% and 25% respectively, of those with BMI greater than 35 kg/m2. Nine percent of patients with BMI less than 18.5 kg/m2 had insulin-controlled diabetes, compared with 23% of those with BMI greater than 35 kg/m2. This table also summarizes the associations of mortality and the individual and overall morbidities with categories of BMI. The left panel in Figure 1 shows the distribution of BMI in our population with a histogram and a smoothed version of the histogram (density estimate). Ninety-five percent of this population had a BMI between 20.0 kg/m2 and 40.1 kg/m2. Figure 1 also shows the increasing trend in BMI of the VA population undergoing CABG procedures during the 14-year period. Note that despite the increase, the median is still contained in the overweight category. Figure 2 contains nonlinear GAM curves, giving the estimated probability of death (left panel) and probability of a serious complication (right panel) across the BMI range, along with 95% pointwise standard error bands, which illustrate the imprecise estimates near the tails. These curves reveal a U-shaped risk function with a minimum around BMI of 30 kg/m2, indicating that patients classified as overweight have the lowest risk, and even those in the lower end of the obese range do not have seriously elevated risk. The U-shape is most pronounced for morbidity, with less of an increase in risk of mortality for the higher BMI range. The estimated risk of operative death for patients with a BMI of 20 kg/m2 (the 2.5th percentile of BMI in our data) is 4.0% ⫾ 0.1% (mean ⫾ standard error of the mean), and at a BMI of 40 kg/m2 (near the 97.5th percentile of BMI in our data) is 3.8% ⫾ 0.1%, compared with the estimated risk at a BMI of 30 kg/m2 (near the 65.7th percentile of BMI in our data) of 3.1% ⫾ 0.1%. Corresponding estimated risks of a serious complication are 13.5% ⫾ 0.2% at a BMI of 20 kg/m2, 13.2% ⫾ 0.2% at a BMI of 40 kg/m2, and 11.7% ⫾ 0.1% at a BMI of 30 kg/m2. Likelihood ratio tests show these nonlinear estimates to be significantly better than linear functions (p ⬍ 0.0001). The linear relationship and the obese category (BMI ⱖ 30 kg/m2) were tested in separate multiple
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
13
Fig 1. The left panel shows the distribution of body mass index (BMI) for 80,792 patients with a histogram and smooth density estimate. The right panel shows the increasing trend in BMI over the 14-year period. The central box shows the data between the quartiles, with the median represented by a line and the whiskers illustrate the extremes (excluding outliers) of the data.
logistic regression models; both were not significant with regard to mortality (p ⫽ 0.533 and 0.734, respectively). Figure 3 displays the relationship between BMI and the odds ratios, using a referent value of BMI of 25 kg/m2, of each outcome for both the univariate and multivariate models. For the multivariate model, the BMI curve represents the estimated relationship to the odds ratio curve in a hypothetical homogeneous risk-matched population. These adjustments account for patterns such as those noted in Table 1, where for example underweight patients are more likely to be smokers and have chronic obstructive pulmonary disease whereas overweight patients are more likely to be younger and diabetic. The minimum of the U-shaped curve of BMI remains in the overweight, slightly obese range regardless of whether none or all of the other covariates are included in the model, indicating that this estimate is robust to imbalances in risk factors across the BMI range. As can be seen from Figure 3, once the other risk factors have been adjusted for, the minimum of the curve shifts slightly to
a lower BMI, and the risk associated with being underweight is lowered while the risk associated with a BMI greater than 35 kg/m2 has increased. The shifting of the curve suggests that for a hypothetical population homogeneous with respect to all the other risk factors, the risk associated with being underweight is decreased while the risk associated with being obese is increased. This may be owing to the fact that the obese patients are younger and the majority of risk variables being adjusted appear to be more closely associated with being underweight. Graphs for the more recent subset of data were identical to those presented. Univariate models were also considered in which the interaction of BMI with the other risk factors was included. The U-shape was surprisingly robust in all of these models, which indicates that even when stratifying by a risk factor this trend is present in all levels of the factor. Figure 4 shows the unadjusted 30-day mortality curves for different levels of age 65 years or older (top panel), current smoking status, diabetes, and overall
Fig 2. The left panel shows the smooth association between the probability (Pr) of 30-day operative mortality and body mass index (BMI) estimated with a generalized additive model. The right panel shows the same association for probability of 30-day occurrence of a major complication. Shaded areas show pointwise standard error bands and the dashed lines indicate the cutoffs for the body mass index categories.
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
14
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Ann Thorac Surg 2007;84:10 –16
Fig 3. Estimated odds ratio (OR) curves of 30-day mortality (left panel) and 30-day morbidity (right panel) from a generalized additive model, using body mass index (BMI) of 25 kg/m2 as the referent value. The dashed lines refer to models with only body mass index. The solid lines refer to models with body mass index and also adjusting for the risk variables mentioned in the methods section.
smoking status (2001 through 2005 only). The density estimates for each of the levels are included below the curves. From the graph illustrating the interaction between age and BMI, it is obvious that the older population has elevated risk compared with the younger popu-
lation, but the shapes of the curves are very similar. The curves for current smokers and nonsmokers differ only slightly. We repeated this analysis using a more complete measure of smoking status that was only collected for the last 4 years in the data set, and the interaction curves for
Fig 4. Probability (Pr) of unadjusted 30-day mortality curves versus body mass index (BMI) for different levels of age (first panel), smoking status (second panel) and diabetes (third panel), and overall smoking status (fourth panel, 2001–2005 only). The density estimates for each of the levels are included below the curves.
this variable are also shown in the last panel of Figure 4. These graphs show that in the range where most of the data lie (eg, BMI between 20 and 40 kg/m2), the risk estimates are very similar between the smoker categories, and because of the smaller sample size the extremes are not well enough estimated to allow conclusions. The three BMI curves for nondiabetic, diet-controlled diabetic, and insulin-controlled diabetic patients in the third panel also retain the U-shape, but there is a slight flattening out of the normal and diet-controlled diabetic curves in the high BMI range, indicating that some of the risk associated with high BMI is attributable to the increased incidence of diabetes in that range. Similar patterns were seen with the morbidity outcome (not shown). Again, the curves for both diabetic groups are based on fewer subjects and so are much less precisely estimated.
Comment The primary goal of this study was to estimate the risks of death and major complications as nonlinear functions of BMI during the period directly after CABG surgery. We found that the relation between BMI and post-CABG outcomes was significantly nonlinear, and when BMI was tested as a linear function or using a category for obesity, it did not result in a significant predictor in the model. Risks were higher in the extremes of BMI, with minimum risk occurring near a BMI of 30 kg/m2. Adjustment for common confounders such as age, diabetes, renal insufficiency, and cardiac disease severity indicated that the U-shape was not attributable to imbalances in these risk variables. The U-shape was also seen in similar forms in analyses stratified by age, smoking, and diabetes. This suggests that intermediary conditions (eg, diabetes) that are traditionally thought to be in the causal pathway between BMI and the outcome of interest or variables that may precede BMI in the causal pathway (eg, smoking resulting in lower BMI) do not account for all of the risk attributable to obesity. The results of this study expand the literature in several ways. First, the results provide evidence that low or high BMI are, indeed, independent risk factors for post-CABG outcomes. This is in contrast to those studies that found no association [10 –12, 14, 15]. The more accurate nonlinear modeling of the association between BMI and post-CABG outcomes resolves previous contradictions in the literature. The U-shape may explain why when left as a continuous variable and estimated linearly, BMI has not provided much information when used in parametric models. For example, in our data set the linear relationship and the obese category were not significant. The standard categorization of BMI also does not seem to describe the relationship very well, as the cutoffs are not located at optimal points along the curve and the curve is not uniform within the categories. A general agreement as to how to model BMI in this literature is needed. The results of this analysis in which the nonlinear association with BMI was clearly demonstrated can aid in this determination. Future modeling of
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
15
BMI should account for this curved relationship. Alternatively, the use of a quadratic BMI term in risk models would be a simple and more accurate way to estimate the BMI curve. Finally, these findings are consistent with studies of BMI in other medical populations that have suggested a U-shaped relationship [1, 5, 13, 15, 22], although the modeling approaches in several of these studies required more stringent assumptions and do not provide as clear a description of the shape of the association. Some previous studies [4, 12, 13] have used stratification on BMI to examine possible related or confounding variables. Although stratification does have some advantages such as simplicity of analysis and interpretation, the statistical methods we have used provide smooth estimates of associations of BMI with outcomes that do not require prespecified cutoffs, provide a closer examination of BMI corresponding to minimum risk, allow for covariate adjustment and interactions, and provide valid statistical inference within these complex situations. In large data sets such as CICSP, the nonlinear methods used in our study can give a more detailed look at complex patterns. Several recent studies have investigated BMI as a preoperative risk factor in CABG surgeries [10 –16]. The conflicting results in these papers have caused controversy as to whether or not BMI is an important risk factor. The results of this paper illustrate that the differences may be owing to the fact that BMI was not optimally modeled. Habib and colleagues [16] use propensity score methods to investigate the effects of body size on outcomes after a CABG procedure. From an analysis viewpoint, both GAM with adjustment for relevant covariates and propensity score methods are options intended to address confounding factors in studies in which randomization is not possible. We have chosen the GAM approach with adjustment for or stratification by other covariates because our primary interest is in estimating the functional shape of the BMI versus mortality or morbidity association. The use of the GAM model for our study provides information on the relationship between BMI and postoperative outcomes, which can be incorporated into the smaller studies that are restricted to certain statistical methods and unable to apply these more robust techniques. There are several issues to consider in the interpretation of this study. First, the study may have limited generalizability given the VA study population (eg, a high proportion of males and high prevalence of comorbid conditions). Second, although the availability of clinical data and robust data set for analysis are strengths of the study, there is always the possibility of unmeasured confounding in observational studies. Finally, in this paper, as in most other published studies, BMI was used to determine obesity. Body mass index may not accurately reflect adiposity in certain types of people [22]. However, for the general population, it is usually assumed that people above a certain BMI have excess fat as well as being overweight [8]. It is also important to note that the present study is cross-sectional and thus does
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
16
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
not take into account recent weight loss and shifts in body weight. In conclusion, obesity has been found to be a major risk factor for many chronic health conditions, and also leads to disability, impairs quality of life, and contributes to escalating health-care costs. Despite these facts, several prior studies reported that obesity is not a risk factor for operative mortality after a CABG procedure. The results of this study show that both very low and very high BMI does affect CABG outcomes, including both mortality and major complications, by means of a nonlinear relationship. Although the lowest risk is associated with those in the overweight range (BMI near 30 kg/m2), there is a significant increase in risk associated with both lower and higher BMI. The results of this study suggest that including BMI in a linear form is not sufficient, and that a more sophisticated modeling approach for BMI should be included in pre-CABG risk assessment and risk adjustment models for CABG surgery outcomes, and support the need for further studies to elucidate the mechanisms of association between both high and low BMI and adverse outcomes, as well as subsequently identifying and evaluating future treatment interventions to reduce risk in these patients.
Ann Thorac Surg 2007;84:10 –16
7.
8. 9.
10. 11. 12.
13.
14. Funding for this study was provided by the VA Central Office, Office of Patient Care Services, to Dr Shroyer for the National Cardiovascular Care Improvement Program as part of the analytical team’s continuous improvement for this national quality assurance program’s risk-modeling endeavors.
15.
16.
References 1. Abrahamowicz M, du Berger R, Grover SA. Flexible modeling of the effects of serum cholesterol on coronary heart disease mortality. Am J Epidemiol 1997;145:714 –29. 2. Allison DB, Fontaine KR, Manson JE, Stevens J, VanItallie TB. Annual deaths attributable to obesity in the United States. JAMA 1999;282:1530 – 8. 3. Durazo-Arvizu R, McGee D, Zhaohai L, Coper R. Establishing the nadir of the body mass index-mortality relationship: a case study. J Am Stat Assoc 1997;92:1312–9. 4. Flegal KM, Graubard BI, Williamson DF, Gail MH. Excess deaths associated with underweight, overweight, and obesity. JAMA 2005;293:1861–7. 5. Gronniger JT. A semiparametric analysis of the relationship of body mass index to mortality. Am J Public Health 2006; 96:173– 8. 6. Gruberg L, Weissman NJ, Waksman R, et al. The impact of obesity on the short-term and long-term outcomes after
17. 18. 19. 20. 21. 22.
percutaneous coronary intervention: the obesity paradox? J Am Coll Cardiol 2002;39:578 – 84. Wessel TR, Arant CB, Olson MB, et al. Relationship of physical fitness versus body mass index with coronary artery disease and cardiovascular events in women. JAMA 2004; 292:1179 – 87. Kuczmarski RJ, Flegal LM. Criteria for definition of overweight in transition: background and recommendations for the United States. Am J Clin Nutr 2000;72:1074 – 81. US Department of Agriculture and US Department of Health and Human Services. Nutrition and your health: dietary guidelines for Americans. Washington, DC: US Government Printing Office, 2000. [Home and Garden Bulletin no. 232.]. Engleman DT, Adams DH, Byrne JG, et al. Impact of body mass index and albumin on morbidity and mortality after cardiac surgery. J Thorac Cariovasc Surg 1999;118:866 –73. Jin R, Grunkemeier GL, Furnary AP, Handy JR Jr. Is obesity a risk factor for mortality in coronary artery bypass surgery? Circulation 2005;111:3359 – 65. Pan W, Hindler K, Lee V, Vaughn W, Collard CD. Obesity in diabetic patients undergoing coronary artery bypass graft surgery is associated with increased postoperative morbidity. Anesthesiology 2006;104:441–7. Prabhakar G, Haan CK, Peterson ED, Coombs LP, Cruzzavala JL, Murray GF. The risks of moderate and extreme obesity for coronary artery bypass grafting outcomes: A study from The Society of Thoracic Surgeons’ Database. Ann Thorac Surg 2002;74:1125–31. Reeves BC, Ascione R, Chamberlain MH, Angelini GD. Effect of body mass index on early outcomes in patients undergoing coronary artery bypass surgery. J Am Coll Cardiol 2003;42:668 –76. Schwann TA, Habib RH, Zacharias A, et al. Effects of body size on operative, intermediate and long-term outcomes after coronary artery bypass operation. Ann Thorac Surg 2001;71:521–31. Habib RH, Zacharias A, Schwann TA, et al. Effects of obesity and small body size on operative and long-term outcomes of coronary artery bypass surgery: a propensity-matched analysis. Ann Thorac Surg 2005;79:1976 – 86. Grover FL, Johnson R, Shroyer LW, Marshall G, Hammermeister KE. The Veterans Affairs Continuous Improvement in Cardiac Surgery study. Ann Thorac Surg 1994;58:1845–51. Grover FL, Shroyer ALW, Hammermeister KE. Calculating risk and outcome: the Veterans Affairs database. Ann Thorac Surg 1996;62(Suppl):6 –11. Fisher SG, Weber L, Goldberg J, Davis F. Mortality ascertainment in the veteran population: alternatives to the National Death Index. Am J Epidemiol 1995;141:242–50. Hastie T, Tibshirani R. Generalized additive models, 1st ed. Boca Raton: Chapman & Hall/CRC, 1990. Schwarz G. Estimating the dimension of a model. Ann Stat 1978;6:461– 4. Allison DB, Faith MS, Heo M, Kotler DP. Hypothesis concerning the U-shape relation between body mass index and mortality. Am J Epidemiol 1997;146:339 – 49.
Relationship of Body Mass Index With Outcomes After Coronary Artery Bypass Graft Surgery Brandie D. Wagner, MS, Gary K. Grunwald, PhD, John S. Rumsfeld, MD, PhD, James O. Hill, PhD, P. Michael Ho, MD, PhD, Holly R. Wyatt, MD, and A. Laurie W. Shroyer, PhD Division of Cardiac Research, Department of Veterans Affairs Medical Center, Eastern Colorado Health Care System; Departments of Preventative Medicine and Biometrics, and Medicine, and Center for Human Nutrition, University of Colorado at Denver and Health Sciences Center, Denver, Colorado
Background. A debate exists whether obesity is a risk factor for operative mortality after coronary artery bypass graft surgery (CABG). The contradictory findings in the literature may largely be attributable to the variety of methodological approaches used to model the association between body mass index (BMI) and post-CABG outcomes. This study aims to investigate this association, and to uncover possible explanations for the lack of consensus across prior studies. Methods. Data were prospectively collected on 80,792 patients who underwent a CABG procedure during a 14-year period at the 45 Department of Veterans Affairs cardiac surgery programs. Generalized additive models were used to estimate the relationship of BMI and outcomes after a CABG procedure. Results. We found that the relationship of BMI with post-CABG mortality and morbidity is U-shaped with the minimum risk located around a BMI of 30 kg/m2,
indicating that patients classified as overweight have the lowest risk, and those in the lower end of the obese range do not have seriously elevated risk. This U-shape relationship is significantly nonlinear and robust to adjustment for other risk factors. Conclusions. This study demonstrates that BMI is an independent predictor of mortality and morbidity after CABG surgery. Previous studies that model BMI linearly or as categories cannot accurately capture this U-shaped relationship and are unlikely to find a significant contribution by including BMI. Further research is needed to determine the mechanisms of risk for patients with low and high BMI and whether interventions to modify BMI may improve patient outcomes.
T
adverse outcomes and prolonged hospitalization after a CABG operation. Thus, it remains unclear whether obesity is an independent risk factor for post-CABG outcomes, and whether obesity is an important target for risk modification to improve outcomes after CABG surgery. A significant limitation of some previous studies examining the relationship between obesity and postCABG outcomes has been the failure to evaluate potential nonlinear associations between body mass index (BMI) and outcomes. In addition, there has been limited formal evaluation of interactions between BMI and key covariates such as age, smoking, and diabetes. Accordingly, the objective of this study was to evaluate, in a large multicenter cohort, the relationship between BMI before surgery with post-CABG outcomes using generalized additive models (GAM). These models provide a flexible method for modeling nonlinear covariate effects, including covariate adjustments and interactions. It is hoped that the results of this study will help clarify the relationship between obesity and post-CABG outcomes, and determine whether BMI is a potential target for interventions to improve patient outcomes for CABG surgery.
he well-documented epidemic of obesity in the United States has led to increased focus on obesity as a risk factor for adverse health outcomes [1–7]. Previous studies have shown that obesity is associated with the development of chronic health conditions including diabetes and coronary artery disease. Moreover, obesity leads to disability and impaired patient quality of life, and contributes to escalating health-care costs [8, 9]. Despite this, controversy remains about the relation between obesity and outcomes among patients with cardiovascular disease. In particular, there is ongoing debate about obesity as a risk factor for adverse outcomes after cardiovascular procedures [6, 10 –16]. Several prior studies have found that obesity is not a risk factor for operative mortality after coronary artery bypass graft (CABG) surgery [10 –12, 14, 15], and it has been suggested that any risk from obesity is directly attributable to clustered risk factors of smoking or diabetes. Other studies [13, 16] have concluded that the extreme obesity category was a significant independent predictor for Accepted for publication March 5, 2007. Address correspondence to Dr Shroyer, Cardiac Research, Denver Department of Veterans Affairs Medical Center, 820 Clermont St (112R), Denver, CO 80220; e-mail: [email protected].
© 2007 by The Society of Thoracic Surgeons Published by Elsevier Inc
(Ann Thorac Surg 2007;84:10 –16) © 2007 by The Society of Thoracic Surgeons
0003-4975/07/$32.00 doi:10.1016/j.athoracsur.2007.03.017
11
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Table 1. Patient Risk Characteristics Risk Variable BMI (mean, SD) Age (mean, SD) Male (%) Current smoker (%) Comorbidities Diabetes (%) Normal Diet-controlled Insulin-controlled Renal insufficiency (%) COPD (%) CVD (%) Cardiac history Prior MI (%) Prior heart surgery (%) Surgical priority (%) Heart failure (%) Angina (%) Outcome variables 30-day mortality (%) 30-day morbidity (%) Endocarditis (%) Renal failure (%) Mediastinitis (%) Bleeding (%) Ventilator (%) RCB (%) Coma (%) Stroke (%) Cardiac arrest (%)
Overall (n ⫽ 80,792)
BMI ⬍ 18.5 (n ⫽ 887)
BMI 18.5–25 (n ⫽ 18,130)
BMI 25–30 (n ⫽ 34,063)
BMI 30–35 (n ⫽ 19,391)
BMI ⬎ 35 (n ⫽ 8,321)
28.6 (5.1) 63.5 (9.3) 99.1 30.4
16.2 (2.1) 64.9 (9.0) 98.9 45.4
22.9 (1.6) 65.0 (9.4) 99.0 38.8
27.4 (1.4) 64.1 (9.2) 99.3 29.6
32.1 (1.4) 62.3 (9.0) 99.1 26.0
38.6 (3.5) 60.4 (8.8) 98.5 24.4
66.8 19.1 14.1 16.2 23.6 20.2
77.0 14.0 9.0 15.9 37.5 25.9
77.7 12.5 9.8 16.8 27.2 23.2
69.5 18.1 12.4 16.3 21.5 20.4
58.6 23.7 17.7 15.8 22.7 18.5
49.7 27.7 22.6 15.2 24.6 16.0
6.9 7.2 5.8 4.3 38.1
9.2 5.3 8.2 5.6 38.9
7.3 7.0 6.8 4.5 40.0
7.1 7.7 5.9 3.8 38.1
6.5 7.4 5.2 4.3 37.0
6.2 5.5 4.7 5.3 36.6
3.4 12.6 ⬍0.1 1.1 1.4 2.7 7.3 0.4 0.7 1.7 2.6
7.1 16.9 0 1.1 1.4 4.5 10.0 0.5 0.7 2.0 4.3
4.0 14.0 ⬍0.1 1.2 1.1 3.8 8.0 0.4 0.8 1.9 3.0
3.2 11.7 ⬍0.1 1.0 1.2 2.6 6.7 0.4 0.6 1.8 2.5
3.0 11.9 ⬍0.1 1.1 1.7 2.0 7.1 0.4 0.7 1.7 2.4
3.6 14.1 ⬍0.1 1.4 2.6 1.9 8.9 0.3 0.7 1.3 2.9
BMI ⫽ body mass index; CABG ⫽ coronary artery bypass graft; COPD ⫽ chronic obstructive pulmonary disease; disease; MI ⫽ myocardial infarction; RCB ⫽ repeat cardiopulmonary bypass; SD ⫽ standard deviation.
Material and Methods The Department of Veterans Affairs (VA) Continuous Improvement in Cardiac Surgery Program (CICSP) prospectively collects patient-level risk, procedural, and outcome data on all patients undergoing cardiac surgery at the 45 VA cardiac surgical centers. The goal of the CICSP is to provide performance reports to VA-based clinical care team members for use in self-assessment and selfimprovement quality of care initiatives. A description of this complete database and the variables captured has been published previously [17, 18]. For this analysis we selected records for all patients undergoing CABG-only procedures from October 1, 1991, to September 30, 2005. From these records, patients with a BMI greater than 58 were excluded owing to the small numbers and inaccurate height recorded for double amputees. No adjustment was made for those with low BMI. There were 67 patients with missing outcome variables and these were excluded, giving a total of 80,792 records. The CICSP project has received both institutional review board ap-
CVD ⫽ cerebrovascular
proval and Health Insurance Portability and Accountability Act waiver for analysis [Colorado Multiple Institutional Review Board protocol 98 – 611]. For this study 12 predictor variables were considered; refer to Table 1 for the list of these variables. All variables are collected within the CICSP and these were chosen a priori for this analysis on the basis of clinical relevance and reliability. Body mass index is a widely used index of obesity and is calculated as weight in kilograms divided by height in meters squared. A dichotomous variable for hypertension, overall smoking status (eg, never smoked versus smoked greater than 3 months or less than 3 months before surgery), and hypoalbuminemia [10] as measured by serum albumin less than 3.5 g/dL were not completely collected within CICSP until 2001 and were therefore included in a subanalysis. The primary outcome measures used in this analysis were 30-day operative mortality and 30-day morbidity (occurrence of at least one of the major complications listed in Table 1). Deaths are verified using the VA Beneficiary Indicator
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
12
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Records Locator System [19], which is a national death registry for all veterans. Because of the on-site data collection used by CICSP’s surgical clinical nurse coordinators, less than 1.2% of the records had any missing values of risk variables. These values were imputed using the median for continuous variables and the most frequent value for categorical variables, which is consistent with the CICSP approach [17, 18]. Hastie and Tibshirani [20] developed the GAM to extend the generalized linear model (including logistic regression) when the form of the relationship between covariates and the outcome is suspected to be nonlinear. Few assumptions are made about the structure of the association of BMI with the outcome, and the data are allowed to shape the functional form of the relationship. These models were fit using the ‘GAM’ function in the S-PLUS software version 6.1 (Insightful Corp, Seattle, WA). Generalized additive models were used in the logistic regression setting to estimate the nonlinear relationship of BMI with both 30-day mortality and morbidity separately. The optimal smoothing parameter for the analysis was chosen based on Bayes information criteria values [21]. As a result, a smoothing spline model with 3 degrees of freedom was used for all smooth model terms. In the preliminary analysis, univariate models containing only preoperative BMI were estimated. To assess the stability of the BMI curve to differences in other patient factors, two further types of analyses were done. First, multivariate logistic regression models (including BMI while adjusting for other risk variables) were estimated. Adjustment for other risk variables accounts for imbalances in risk factors across the range of BMI (eg, more smokers in the low BMI range or diabetics in the high BMI range). In these models, age and surgery year were included in nonlinear form. Second, models including interactions between BMI and other patient risk factors were estimated. These allow different nonlinear association patterns between BMI and the outcome for different risk groups (eg, smokers versus nonsmokers or diabetics versus nondiabetics). These models were fit by estimating a separate BMI curve for each level of the categorical risk variable. These analyses were then repeated using a more recent subset of the data (2001 to 2005), which included additional risk variables that were of interest for adjustment. These variables include hypertension, overall smoking status, and hypoalbuminemia. Using a semiparametric GAM, the estimated effect of BMI as a continuous risk factor is represented by a flexible function rather than a single parameter, so the best way to display the results is to plot the function. Pointwise standard error bands are used to show the precision of the estimated curve. To test the significance of the nonlinear effects, a likelihood ratio test was used. Significant results of this test indicate that the nonparametric smoothing spline fits the data significantly better than a linear function.
Ann Thorac Surg 2007;84:10 –16
Results The distributions across BMI classes of these risk factors for the patients included in this analysis are displayed in Table 1. Of the 80,792 patients, 887 (1.1%) were classified as underweight (BMI ⬍ 18.5), 18,130 (22.4%) were normal weight (18.5 ⱕ BMI ⬍ 25), 34,063 (42.2%) were overweight (25 ⱕ BMI ⬍ 30), 19,391 (24.0%) were obese (30 ⱕ BMI ⬍ 35), and 8,321 (10.3%) were morbidly obese (BMI ⱖ 35). Increasing BMI was associated with younger age, lower rates of smoking and chronic obstructive pulmonary disease (chronic obstructive pulmonary disease does increase slightly in the higher BMI categories), lower rates of previous myocardial infarction and cerebral vascular disease, lower rates of emergent surgical priority, and higher rates of insulin-controlled diabetes. In particular, for patients with BMI less than 18.5 kg/m2, 45% were current smokers and 38% had chronic obstructive pulmonary disease compared with 24% and 25% respectively, of those with BMI greater than 35 kg/m2. Nine percent of patients with BMI less than 18.5 kg/m2 had insulin-controlled diabetes, compared with 23% of those with BMI greater than 35 kg/m2. This table also summarizes the associations of mortality and the individual and overall morbidities with categories of BMI. The left panel in Figure 1 shows the distribution of BMI in our population with a histogram and a smoothed version of the histogram (density estimate). Ninety-five percent of this population had a BMI between 20.0 kg/m2 and 40.1 kg/m2. Figure 1 also shows the increasing trend in BMI of the VA population undergoing CABG procedures during the 14-year period. Note that despite the increase, the median is still contained in the overweight category. Figure 2 contains nonlinear GAM curves, giving the estimated probability of death (left panel) and probability of a serious complication (right panel) across the BMI range, along with 95% pointwise standard error bands, which illustrate the imprecise estimates near the tails. These curves reveal a U-shaped risk function with a minimum around BMI of 30 kg/m2, indicating that patients classified as overweight have the lowest risk, and even those in the lower end of the obese range do not have seriously elevated risk. The U-shape is most pronounced for morbidity, with less of an increase in risk of mortality for the higher BMI range. The estimated risk of operative death for patients with a BMI of 20 kg/m2 (the 2.5th percentile of BMI in our data) is 4.0% ⫾ 0.1% (mean ⫾ standard error of the mean), and at a BMI of 40 kg/m2 (near the 97.5th percentile of BMI in our data) is 3.8% ⫾ 0.1%, compared with the estimated risk at a BMI of 30 kg/m2 (near the 65.7th percentile of BMI in our data) of 3.1% ⫾ 0.1%. Corresponding estimated risks of a serious complication are 13.5% ⫾ 0.2% at a BMI of 20 kg/m2, 13.2% ⫾ 0.2% at a BMI of 40 kg/m2, and 11.7% ⫾ 0.1% at a BMI of 30 kg/m2. Likelihood ratio tests show these nonlinear estimates to be significantly better than linear functions (p ⬍ 0.0001). The linear relationship and the obese category (BMI ⱖ 30 kg/m2) were tested in separate multiple
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
13
Fig 1. The left panel shows the distribution of body mass index (BMI) for 80,792 patients with a histogram and smooth density estimate. The right panel shows the increasing trend in BMI over the 14-year period. The central box shows the data between the quartiles, with the median represented by a line and the whiskers illustrate the extremes (excluding outliers) of the data.
logistic regression models; both were not significant with regard to mortality (p ⫽ 0.533 and 0.734, respectively). Figure 3 displays the relationship between BMI and the odds ratios, using a referent value of BMI of 25 kg/m2, of each outcome for both the univariate and multivariate models. For the multivariate model, the BMI curve represents the estimated relationship to the odds ratio curve in a hypothetical homogeneous risk-matched population. These adjustments account for patterns such as those noted in Table 1, where for example underweight patients are more likely to be smokers and have chronic obstructive pulmonary disease whereas overweight patients are more likely to be younger and diabetic. The minimum of the U-shaped curve of BMI remains in the overweight, slightly obese range regardless of whether none or all of the other covariates are included in the model, indicating that this estimate is robust to imbalances in risk factors across the BMI range. As can be seen from Figure 3, once the other risk factors have been adjusted for, the minimum of the curve shifts slightly to
a lower BMI, and the risk associated with being underweight is lowered while the risk associated with a BMI greater than 35 kg/m2 has increased. The shifting of the curve suggests that for a hypothetical population homogeneous with respect to all the other risk factors, the risk associated with being underweight is decreased while the risk associated with being obese is increased. This may be owing to the fact that the obese patients are younger and the majority of risk variables being adjusted appear to be more closely associated with being underweight. Graphs for the more recent subset of data were identical to those presented. Univariate models were also considered in which the interaction of BMI with the other risk factors was included. The U-shape was surprisingly robust in all of these models, which indicates that even when stratifying by a risk factor this trend is present in all levels of the factor. Figure 4 shows the unadjusted 30-day mortality curves for different levels of age 65 years or older (top panel), current smoking status, diabetes, and overall
Fig 2. The left panel shows the smooth association between the probability (Pr) of 30-day operative mortality and body mass index (BMI) estimated with a generalized additive model. The right panel shows the same association for probability of 30-day occurrence of a major complication. Shaded areas show pointwise standard error bands and the dashed lines indicate the cutoffs for the body mass index categories.
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
14
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
Ann Thorac Surg 2007;84:10 –16
Fig 3. Estimated odds ratio (OR) curves of 30-day mortality (left panel) and 30-day morbidity (right panel) from a generalized additive model, using body mass index (BMI) of 25 kg/m2 as the referent value. The dashed lines refer to models with only body mass index. The solid lines refer to models with body mass index and also adjusting for the risk variables mentioned in the methods section.
smoking status (2001 through 2005 only). The density estimates for each of the levels are included below the curves. From the graph illustrating the interaction between age and BMI, it is obvious that the older population has elevated risk compared with the younger popu-
lation, but the shapes of the curves are very similar. The curves for current smokers and nonsmokers differ only slightly. We repeated this analysis using a more complete measure of smoking status that was only collected for the last 4 years in the data set, and the interaction curves for
Fig 4. Probability (Pr) of unadjusted 30-day mortality curves versus body mass index (BMI) for different levels of age (first panel), smoking status (second panel) and diabetes (third panel), and overall smoking status (fourth panel, 2001–2005 only). The density estimates for each of the levels are included below the curves.
this variable are also shown in the last panel of Figure 4. These graphs show that in the range where most of the data lie (eg, BMI between 20 and 40 kg/m2), the risk estimates are very similar between the smoker categories, and because of the smaller sample size the extremes are not well enough estimated to allow conclusions. The three BMI curves for nondiabetic, diet-controlled diabetic, and insulin-controlled diabetic patients in the third panel also retain the U-shape, but there is a slight flattening out of the normal and diet-controlled diabetic curves in the high BMI range, indicating that some of the risk associated with high BMI is attributable to the increased incidence of diabetes in that range. Similar patterns were seen with the morbidity outcome (not shown). Again, the curves for both diabetic groups are based on fewer subjects and so are much less precisely estimated.
Comment The primary goal of this study was to estimate the risks of death and major complications as nonlinear functions of BMI during the period directly after CABG surgery. We found that the relation between BMI and post-CABG outcomes was significantly nonlinear, and when BMI was tested as a linear function or using a category for obesity, it did not result in a significant predictor in the model. Risks were higher in the extremes of BMI, with minimum risk occurring near a BMI of 30 kg/m2. Adjustment for common confounders such as age, diabetes, renal insufficiency, and cardiac disease severity indicated that the U-shape was not attributable to imbalances in these risk variables. The U-shape was also seen in similar forms in analyses stratified by age, smoking, and diabetes. This suggests that intermediary conditions (eg, diabetes) that are traditionally thought to be in the causal pathway between BMI and the outcome of interest or variables that may precede BMI in the causal pathway (eg, smoking resulting in lower BMI) do not account for all of the risk attributable to obesity. The results of this study expand the literature in several ways. First, the results provide evidence that low or high BMI are, indeed, independent risk factors for post-CABG outcomes. This is in contrast to those studies that found no association [10 –12, 14, 15]. The more accurate nonlinear modeling of the association between BMI and post-CABG outcomes resolves previous contradictions in the literature. The U-shape may explain why when left as a continuous variable and estimated linearly, BMI has not provided much information when used in parametric models. For example, in our data set the linear relationship and the obese category were not significant. The standard categorization of BMI also does not seem to describe the relationship very well, as the cutoffs are not located at optimal points along the curve and the curve is not uniform within the categories. A general agreement as to how to model BMI in this literature is needed. The results of this analysis in which the nonlinear association with BMI was clearly demonstrated can aid in this determination. Future modeling of
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
15
BMI should account for this curved relationship. Alternatively, the use of a quadratic BMI term in risk models would be a simple and more accurate way to estimate the BMI curve. Finally, these findings are consistent with studies of BMI in other medical populations that have suggested a U-shaped relationship [1, 5, 13, 15, 22], although the modeling approaches in several of these studies required more stringent assumptions and do not provide as clear a description of the shape of the association. Some previous studies [4, 12, 13] have used stratification on BMI to examine possible related or confounding variables. Although stratification does have some advantages such as simplicity of analysis and interpretation, the statistical methods we have used provide smooth estimates of associations of BMI with outcomes that do not require prespecified cutoffs, provide a closer examination of BMI corresponding to minimum risk, allow for covariate adjustment and interactions, and provide valid statistical inference within these complex situations. In large data sets such as CICSP, the nonlinear methods used in our study can give a more detailed look at complex patterns. Several recent studies have investigated BMI as a preoperative risk factor in CABG surgeries [10 –16]. The conflicting results in these papers have caused controversy as to whether or not BMI is an important risk factor. The results of this paper illustrate that the differences may be owing to the fact that BMI was not optimally modeled. Habib and colleagues [16] use propensity score methods to investigate the effects of body size on outcomes after a CABG procedure. From an analysis viewpoint, both GAM with adjustment for relevant covariates and propensity score methods are options intended to address confounding factors in studies in which randomization is not possible. We have chosen the GAM approach with adjustment for or stratification by other covariates because our primary interest is in estimating the functional shape of the BMI versus mortality or morbidity association. The use of the GAM model for our study provides information on the relationship between BMI and postoperative outcomes, which can be incorporated into the smaller studies that are restricted to certain statistical methods and unable to apply these more robust techniques. There are several issues to consider in the interpretation of this study. First, the study may have limited generalizability given the VA study population (eg, a high proportion of males and high prevalence of comorbid conditions). Second, although the availability of clinical data and robust data set for analysis are strengths of the study, there is always the possibility of unmeasured confounding in observational studies. Finally, in this paper, as in most other published studies, BMI was used to determine obesity. Body mass index may not accurately reflect adiposity in certain types of people [22]. However, for the general population, it is usually assumed that people above a certain BMI have excess fat as well as being overweight [8]. It is also important to note that the present study is cross-sectional and thus does
CARDIOVASCULAR
Ann Thorac Surg 2007;84:10 –16
CARDIOVASCULAR
16
WAGNER ET AL RELATING BMI AND POST-CABG OUTCOMES
not take into account recent weight loss and shifts in body weight. In conclusion, obesity has been found to be a major risk factor for many chronic health conditions, and also leads to disability, impairs quality of life, and contributes to escalating health-care costs. Despite these facts, several prior studies reported that obesity is not a risk factor for operative mortality after a CABG procedure. The results of this study show that both very low and very high BMI does affect CABG outcomes, including both mortality and major complications, by means of a nonlinear relationship. Although the lowest risk is associated with those in the overweight range (BMI near 30 kg/m2), there is a significant increase in risk associated with both lower and higher BMI. The results of this study suggest that including BMI in a linear form is not sufficient, and that a more sophisticated modeling approach for BMI should be included in pre-CABG risk assessment and risk adjustment models for CABG surgery outcomes, and support the need for further studies to elucidate the mechanisms of association between both high and low BMI and adverse outcomes, as well as subsequently identifying and evaluating future treatment interventions to reduce risk in these patients.
Ann Thorac Surg 2007;84:10 –16
7.
8. 9.
10. 11. 12.
13.
14. Funding for this study was provided by the VA Central Office, Office of Patient Care Services, to Dr Shroyer for the National Cardiovascular Care Improvement Program as part of the analytical team’s continuous improvement for this national quality assurance program’s risk-modeling endeavors.
15.
16.
References 1. Abrahamowicz M, du Berger R, Grover SA. Flexible modeling of the effects of serum cholesterol on coronary heart disease mortality. Am J Epidemiol 1997;145:714 –29. 2. Allison DB, Fontaine KR, Manson JE, Stevens J, VanItallie TB. Annual deaths attributable to obesity in the United States. JAMA 1999;282:1530 – 8. 3. Durazo-Arvizu R, McGee D, Zhaohai L, Coper R. Establishing the nadir of the body mass index-mortality relationship: a case study. J Am Stat Assoc 1997;92:1312–9. 4. Flegal KM, Graubard BI, Williamson DF, Gail MH. Excess deaths associated with underweight, overweight, and obesity. JAMA 2005;293:1861–7. 5. Gronniger JT. A semiparametric analysis of the relationship of body mass index to mortality. Am J Public Health 2006; 96:173– 8. 6. Gruberg L, Weissman NJ, Waksman R, et al. The impact of obesity on the short-term and long-term outcomes after
17. 18. 19. 20. 21. 22.
percutaneous coronary intervention: the obesity paradox? J Am Coll Cardiol 2002;39:578 – 84. Wessel TR, Arant CB, Olson MB, et al. Relationship of physical fitness versus body mass index with coronary artery disease and cardiovascular events in women. JAMA 2004; 292:1179 – 87. Kuczmarski RJ, Flegal LM. Criteria for definition of overweight in transition: background and recommendations for the United States. Am J Clin Nutr 2000;72:1074 – 81. US Department of Agriculture and US Department of Health and Human Services. Nutrition and your health: dietary guidelines for Americans. Washington, DC: US Government Printing Office, 2000. [Home and Garden Bulletin no. 232.]. Engleman DT, Adams DH, Byrne JG, et al. Impact of body mass index and albumin on morbidity and mortality after cardiac surgery. J Thorac Cariovasc Surg 1999;118:866 –73. Jin R, Grunkemeier GL, Furnary AP, Handy JR Jr. Is obesity a risk factor for mortality in coronary artery bypass surgery? Circulation 2005;111:3359 – 65. Pan W, Hindler K, Lee V, Vaughn W, Collard CD. Obesity in diabetic patients undergoing coronary artery bypass graft surgery is associated with increased postoperative morbidity. Anesthesiology 2006;104:441–7. Prabhakar G, Haan CK, Peterson ED, Coombs LP, Cruzzavala JL, Murray GF. The risks of moderate and extreme obesity for coronary artery bypass grafting outcomes: A study from The Society of Thoracic Surgeons’ Database. Ann Thorac Surg 2002;74:1125–31. Reeves BC, Ascione R, Chamberlain MH, Angelini GD. Effect of body mass index on early outcomes in patients undergoing coronary artery bypass surgery. J Am Coll Cardiol 2003;42:668 –76. Schwann TA, Habib RH, Zacharias A, et al. Effects of body size on operative, intermediate and long-term outcomes after coronary artery bypass operation. Ann Thorac Surg 2001;71:521–31. Habib RH, Zacharias A, Schwann TA, et al. Effects of obesity and small body size on operative and long-term outcomes of coronary artery bypass surgery: a propensity-matched analysis. Ann Thorac Surg 2005;79:1976 – 86. Grover FL, Johnson R, Shroyer LW, Marshall G, Hammermeister KE. The Veterans Affairs Continuous Improvement in Cardiac Surgery study. Ann Thorac Surg 1994;58:1845–51. Grover FL, Shroyer ALW, Hammermeister KE. Calculating risk and outcome: the Veterans Affairs database. Ann Thorac Surg 1996;62(Suppl):6 –11. Fisher SG, Weber L, Goldberg J, Davis F. Mortality ascertainment in the veteran population: alternatives to the National Death Index. Am J Epidemiol 1995;141:242–50. Hastie T, Tibshirani R. Generalized additive models, 1st ed. Boca Raton: Chapman & Hall/CRC, 1990. Schwarz G. Estimating the dimension of a model. Ann Stat 1978;6:461– 4. Allison DB, Faith MS, Heo M, Kotler DP. Hypothesis concerning the U-shape relation between body mass index and mortality. Am J Epidemiol 1997;146:339 – 49.