- Email: [email protected]
Outcomes after esophagectomy: a ten-year prospective cohort
J. MAXWELL CHAMBERLAIN MEMORIAL PAPER
Outcomes After Esophagectomy: A Ten-Year Prospective Cohort Stephen H. Bailey, MD, David A. Bull, MD, David H. Harpole, MD, Jeffrey J. Rentz, MD, Leigh A. Neumayer, MD, Theodore N. Pappas, MD, Jennifer Daley, MD, William G. Henderson, PhD, Barbara Krasnicka, PhD, and Shukri F. Khuri, MD Veterans Affairs Medical Center, University of Utah Medical School, Salt Lake City, Utah; Veterans Affairs Medical Center, Duke University Medical School, Durham, North Carolina; Institute for Health Policy, Massachusetts General Hospital/Partners Healthcare System and Department of Medicine, Harvard Medical School, Boston, Massachusetts; Veterans Affairs Medical Center, Harvard Medical School, Brockton/West Roxbury, Massachusetts; and Cooperative Studies Program Coordinating Center, Veterans Affairs Medical Center, Hines, Illinois
Background. The Department of Veterans Affairs National Surgical Quality Improvement Program is a unique resource to prospectively analyze surgical outcomes from a cross-section of surgical services nationally. We used this database to assess risk factors for morbidity and mortality after esophagectomy in Veterans Affairs Medical Centers from 1991 to 2001. Methods. A total of 1,777 patients underwent an esophagectomy at 109 Veterans Affairs hospitals with complete in-hospital and 30-day outcomes recorded. Bivariate and multivariable analyses were completed. Results. Thirty-day mortality was 9.8% (174/1,777) and the incidence of one or more of 20 predefined complications was 49.5% (880/1,777). The most frequent postoperative complications were pneumonia in 21% (380/1,777), respiratory failure in 16% (288/1,777), and ventilator support more than 48 hours in 22% (387/1,777). Preoperative predictors of mortality based on multivariable analysis included neoadjuvant therapy, blood urea nitrogen level
of more than 40 mg/dL, alkaline phosphatase level of more than 125 U/L, diabetes mellitus, alcohol abuse, decreased functional status, ascites, and increasing age. Preoperative factors impacting morbidity were increasing age, dyspnea, diabetes mellitus, chronic obstructive pulmonary disease, alkaline phosphatase level of more than 125 U/L, lower serum albumin concentration, increased complexity score, and decreased functional status. Intraoperative risk factors for mortality included the need for transfusion; intraoperative risk factors for morbidity included the need for transfusion and longer operative time. Conclusions. These data constitute the largest prospective outcomes cohort in the literature and document a near 50% morbidity rate and 10% mortality rate after esophagectomy. Data from this study can be used to better stratify patients before esophagectomy. (Ann Thorac Surg 2003;75:217–22) © 2003 by The Society of Thoracic Surgeons
E
models of morbidity and mortality allowing comparative assessments of quality of care among institutions nationwide.
sophageal resection is associated with high rates of perioperative morbidity and mortality. Most series reporting outcomes after esophagectomy are from single institutions and have small numbers of patients with varying comorbidities. These drawbacks make it difficult to extrapolate the results of these series to the general population of patients undergoing esophageal resection nationally [1– 8]. As a result, defining risk factors associated with adverse perioperative outcomes is problematic. The Department of Veterans Affairs (VA) National Surgical Quality Improvement Program (NSQIP) was designed to reliably and prospectively collect pertinent historical, laboratory, intraoperative, and patient data for surgical procedures covering a number of specialties [9 –13]. The goal of the NSQIP is to develop risk-adjusted Presented at the Thirty-eighth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 28 –30, 2002. Address reprint requests to Dr Bull, 50 North Medical Drive, University of Utah, Division of Cardiothoracic Surgery, Room 3C127, Salt Lake City, UT 84132; e-mail: [email protected].
© 2003 by The Society of Thoracic Surgeons Published by Elsevier Science Inc
Patients and Methods The methods of the VA NSQIP have been described in detail elsewhere [9 –11].
Population Data were acquired prospectively from 109 Department of Veterans Affairs Medical Centers performing esophagectomy between January 1991 and December 2000. The database included 1,777 patients who had undergone esophageal resection (current procedural terminology codes 43107 to 43124). Esophagectomy was performed for malignancy in 84.9% (n ⫽ 1,509) of cases and benign disease in 15.1% (n ⫽ 268). The number of procedures per institution ranged from 1 to 55 and 33 institutions performed 20 or more resections. 0003-4975/03/$30.00 PII S0003-4975(02)04368-0
218
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Patient Characteristics A dedicated, trained nurse reviewer recorded 122 individual variables for each operation. Each reviewer was trained and continually tested and monitored on a set of standard criteria and definitions for coding all collected variables. These variables were selected by an expert advisory panel and from literature reviews for the NSQIP. Most laboratory values and demographic variables were downloaded directly to the statistical center from computers at each participating VA Medical Center. A series of data audits were performed on an ongoing basis to ensure data completeness and reliability. Each center underwent periodic internal and random external audits for data quality. A more detailed description of the auditing process is available elsewhere [11].
Statistical Analysis Bivariate analyses were performed relating the demographic, laboratory, and preoperative variables to 30-day morbidity and mortality. The unpaired t test was used for continuous variables and the 2 test was used for categorical variables. Variables with a prevalence more than 0.5% and that were significant on bivariate analysis at p ⬍ 0.20 were considered potential independent variables in a multivariable logistic regression. Stepwise logistic regression analysis with entry and exit criteria set at p ⱕ 0.05 was used with mortality as the dependent variable. For morbidity analyses, the presence or absence of at least one complication was considered the dependent variable. The multivariable regression model for 30-day mortality and morbidity was repeated, incorporating intraoperative variables such as operative time and blood loss.
Results Patient Characteristics and Outcomes The study population consisted of 1,777 patients who underwent esophagectomy. The sample was predominantly male (99.1%, n ⫽ 1,761) with a mean age of 63.4 ⫾ 9.9 years. Esophageal resection was performed for a neoplastic process in 84.9% (n ⫽ 1509) and for benign disease in 15.1% (n ⫽ 269). Other patient characteristics are listed in Table 1. Thirty-day mortality was 9.8% (n ⫽ 174) and morbidity was 49.5% (n ⫽ 880). No difference was noted in 30-day mortality when comparing patients with benign disease (10.1%) and patients with malignancy (9.7%). Likewise, 30-day morbidity was not different between patients with benign disease (51.1%) and patients with malignancy (49.0%). The most common complications were pulmonary: pneumonia in 21.4% (n ⫽ 380), ventilator dependence for more than 48 hours in 21.8% (n ⫽ 387), and unplanned reintubation in 16.2% (n ⫽ 288). Other complications are listed in Table 2.
Bivariate Analyses Patient characteristics associated with an increased risk of 30-day mortality (p ⬍ 0.2) included 21 preoperative clinical variables, 10 laboratory values, and 5 periopera-
Ann Thorac Surg 2003;75:217–22
Table 1. Patient Characteristics and Preoperative Risk Factors Variable Male Smoking within 2 weeks ⬎ 10% weight loss Severe COPD More than 2 drinks/day within 2 weeks Dyspnea at rest/minimal exertion Diabetes Partially or totally dependent Disseminated cancer History of CVA with deficit Chronic steroid use Congestive Heart Failure Impaired sensorium Hemiplegia Ascites COPD ⫽ chronic obstructive pulmonary disease; cular accident.
n/N
%
1761/1777 784/1749 569/1777 402/1777 341/1742 341/1759 221/1749 148/1777 142/1777 70/1749 46/1777 35/1777 36/1777 26/1749 10/1749
99.1 44.8 32.0 22.6 19.6 19.4 12.6 8.3 8.0 4.0 2.6 2.0 2.0 1.5 0.6
CVA ⫽ cerebrovas-
tive variables. Fifteen preoperative variables, 12 laboratory values, and 7 perioperative variables were associated with increased risk of 30-day morbidity.
Multivariable Analyses All factors with a prevalence of greater than 0.5% and a bivariate significance of p ⬍ 0.2 were included in multivariable stepwise logistic regression analysis to identify independent risk factors for morbidity and mortality. Preoperative independent predictors of mortality (p ⬍ 0.05) included neoadjuvant therapy, insulin-dependent diabetes mellitus, current alcohol use, decreased functional status, ascites, blood urea nitrogen (BUN) level of more than 40 mg/dL, alkaline phosphatase level of more than 125 U/L, and increasing age. The -coefficients for these variables as well as their p values and odds ratios are found in Table 3. When the model was repeated with the inclusion of intraoperative variables, all of the preoperative factors continued to be independently associated with mortality except for alkaline phosphatase concentration. In this model the need for intraoperative blood transfusion was also predictive of mortality. The results of the multivariable analysis including preoperative and intraoperative factors are shown in Table 4. Each model was then repeated with malignancy added as a separate variable. The presence of malignancy was not associated with an increased risk of 30-day morbidity or mortality (data not shown). Preoperative independent risk factors for overall morbidity (p ⬍ 0.05) on multivariable analysis included increasing age, the presence of dyspnea with mild exertion, insulin-dependent diabetes mellitus, chronic obstructive pulmonary disease (COPD), current smoking, alkaline phosphatase level more than 125 U/L, decreased functional status, lower albumin and higher complexity score. The -coefficient, standard error, p-value and odds ratio are illustrated in Table 5. When this multivariable
Ann Thorac Surg 2003;75:217–22
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Table 2. Thirty-Day Morbidity and Mortality Variable Thirty-day mortality Thirty-day morbidity Wound complications Superficial Deep Dehiscence Respiratory complications Pneumonia Reintubation Pulmonary embolism Failure to wean ⬎ 48 h Renal complications Progressive renal insufficiency Acute renal failure Urinary tract infection CNS complications CVA Coma ⬎ 24 h Peripheral nerve injury Cardiac complications Cardiac arrest Pulmonary edema Myocardial infarction Other complications Prolonged ileus Bleeding requiring ⬎ 4 units PRBC Graft failure DVT Systemic sepsis
n
%
174 880
9.8 49.5
95 100 66
5.3 5.6 3.7
380 288 13 387
21.4 16.2 0.7 21.8
49 37 89
2.8 2.1 5.0
6 11 6
0.3 0.6 0.3
98 64 21
5.5 7.6 1.2
76 98 14 16 169
4.3 5.5 0.8 0.9 9.5
219
Table 4. Multivariable Logistic Regression Model of Preoperative and Intraoperative Variables Predictive of 30Day Mortality Variable
 coefficient
SE
p
OR
⫺5.87 0.05 0.15 2.75 0.43 1.37 0.50 0.47 0.53
0.67 0.01 0.03 0.75 0.13 0.50 0.20 0.18 0.23
0.0001 0.0001 0.0001 0.0002 0.001 0.006 0.01 0.01 0.02
1.05 1.17 15.7 1.53 3.94 1.65 1.60 1.69
Intercept Age Intraop. RBC Ascites Diabetes BUN Alcohol Functional status Neo. therapy ⫺2 Log likelihood ⫽ 1139; statistic ⫽ 8.34 (p ⫽ 0.40).
C-index ⫽ 0.71;
Hosmer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis. BUN ⫽ blood urea nitrogen ⬎ 40 mg/dL; Intraop. RBC ⫽ blood transfusion during surgery; Neo. therapy ⫽ neoadjuvant chemotherapy or radiation therapy; OR ⫽ odds ratio; SE ⫽ standard error.
age, dyspnea, diabetes mellitus, COPD, complexity score, and alkaline phosphatase level of more than 125 U/L (Table 6). The significant intraoperative variables predictive of morbidity included the need for intraoperative blood transfusion, longer operative time, and emergency status.
Model Validation
analysis was repeated incorporating intraoperative variables, preoperative variables that continued to be independently associated with morbidity included increasing
The predictive validity of these models was measured by the c-index. The c-index for the mortality analysis was 0.69 for the model incorporating only preoperative factors and 0.71 when preoperative and intraoperative variables were included. The c-index was 0.62 for the morbidity model when only preoperative variables were included and 0.65 for the model including preoperative and intraoperative variables. These c-indices indicate a moderate level of predictability. The model fit was mea-
Table 3. Multivariable Logistic Regression Model of Preoperative Variables Predictive of 30-Day Mortality
Table 5. Multivariable Logistic Regression Model Including Preoperative Variables Predictive of 30-Day Morbidity
CNS ⫽ central nervous system; CVA ⫽ cerebrovascular accident; DVT ⫽ deep vein thrombosis; PRBC ⫽ packed red blood cells.
Variable
 coefficient
SE
p
OR
⫺5.65 0.05 3.03 0.42 0.48 0.62 1.30 0.48 0.45
0.66 0.01 0.74 0.13 0.18 0.22 0.50 0.20 0.22
0.001 0.0001 0.0001 0.002 0.007 0.01 0.01 0.02 0.05
1.05 20.66 1.52 1.62 1.85 3.67 1.62 1.56
Intercept Age Ascites Diabetes Functional status Neo. therapy BUN Alcohol Alk phos ⫺2 Log likelihood ⫽ 1139; statistic ⫽ 3.01 (p ⫽ 0.93).
C-index ⫽ 0.69;
Hosmer-Lemeshow
Variable
 coefficient
SE
p
OR
⫺1.31 0.38 0.31 0.11 ⫺0.21 0.30 0.36 0.01 0.29
0.34 0.10 0.12 0.04 0.09 0.13 0.15 0.005 0.15
0.0001 0.0001 0.01 0.01 0.02 0.02 0.02 0.03 0.05
1.47 1.36 1.11 0.81 1.36 1.43 1.01 1.34
Intercept Diabetes Dyspnea Complexity score Albumin COPD Alk phos Age Functional status ⫺2 Log likelihood ⫽ 2463; statistic ⫽ 11.20 (p ⫽ 0.19).
C-index ⫽ 0.62;
Homer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis.
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis.
Alk phos ⫽ serum alkaline phosphatase ⬎ 125 U/L; BUN ⫽ blood urea nitrogen ⬎ 40 mg/dL; Neo. therapy ⫽ Neoadjuvant chemotherapy or radiation therapy; OR ⫽ odds ratio; SE ⫽ standard error.
Alk phos ⫽ serum alkaline phosphatase ⬎ 125 U/L; obstructive pulmonary disease; OR ⫽ odds ratio; error.
COPD ⫽ chronic SE ⫽ standard
220
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Ann Thorac Surg 2003;75:217–22
Table 6. Multivariable Regression Models Including Preoperative and Intraoperative Variables Predictive of 30Day Morbidity Variable
 coefficient
SE
p
OR
⫺1.88 0.15 0.38 0.36 0.30 0.01 0.06 0.10 0.37 0.72
0.38 0.03 0.10 0.13 0.12 0.005 0.02 0.04 0.15 0.31
0.0001 0.0001 0.0002 0.004 0.01 0.02 0.02 0.02 0.02 0.02
1.16 1.46 1.44 1.36 1.01 1.06 1.10 1.44 2.85
Intercept Intraop. PRBC Diabetes COPD Dyspnea Age Operative time Complexity score Alk phos Emergency ⫺2 Log likelihood ⫽ 2463; statistic ⫽ 6.40 (p ⫽ 0.60).
C-index ⫽ 0.65;
Homer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis. Alk phos ⫽ serum alkaline phosphatase ⬎ 125 mg/dL; Intraop. PRBC ⫽ intraoperative packed red blood cells (blood transfusion during surgery); OR ⫽ odds ratio; SE ⫽ standard error.
sured with the Hosmer-Lemeshow statistic. The HosmerLemeshow statistic for the mortality model incorporating preoperative variables was 3.01 (p ⫽ 0.93) and for the mortality model incorporating preoperative and intraoperative variables was 8.34 (p ⫽ 0.40). The HosmerLemeshow statistic was 11.20 (p ⫽ 0.19) for the morbidity model incorporating only preoperative variables and for the morbidity model including preoperative and intraoperative variables was 6.40 (p ⫽ 0.60). These HosmerLemeshow statistics indicate a good level of fit for each model.
Comment In this study, we analyzed the factors affecting perioperative morbidity and mortality in 1,777 patients undergoing esophagectomy in 109 VA Medical Centers during a 10-year period. This report represents the largest prospective study of perioperative outcomes after esophagectomy. The VA NSQIP offers a unique opportunity to study perioperative outcomes because of the large number of patients enrolled and the reliable prospective collection of data by trained and audited nurses. Our analysis demonstrated a 50% perioperative morbidity and 10% perioperative mortality after esophagectomy. These numbers are within the range of recent results in the published literature. High-volume singleinstitution series trend toward the lowest incidence of adverse perioperative events. Orringer and colleagues [3] reported a 4% mortality and approximately 26% morbidity in a series of 1,085 patients undergoing transhiatal esophagectomy. Hagen and associates [8] reported a perioperative mortality of 6% in a series of 100 transthoracic esophagectomies for esophageal adenocarcinoma. Altorki and Skinner [6] demonstrated a 49% morbidity and 5% mortality in a series of 111 patients who underwent transthoracic esophagectomy with radical lymph-
adenectomy for esophageal malignancy. Boyle and colleagues [7] reported a series including both transhiatal and transthoracic esophagectomies with perioperative morbidity of 60% and mortality of 12%. A number of factors contributed to the morbidity and mortality observed in this study. Comorbidity was common among the study patients. The most common comorbidities were active smoking (44%), alcohol abuse (19%), dyspnea at rest or with minimal exertion (19%), severe COPD (22%), chronic steroid use (10%), and disseminated cancer (8%). Data exist to support the hypothesis that patients cared for in VA Medical Centers are indeed more ill at base line than the general population [14]. Further, most prior studies in the literature reflect outcomes after elective operations. By contrast, our series included 54 patients in whom the procedure was performed under emergent circumstances. Relatively few studies in the literature have used multivariable analysis to evaluate risk factors for adverse perioperative outcomes after esophagectomy. These studies have consistently demonstrated that decreased functional status is a significant predictor of poor outcomes after esophagectomy [15–17]. Advanced patient age has been shown to be associated with an increased risk of perioperative morbidity and mortality after esophagectomy [15, 17]. Finally, the presence of diabetes mellitus in patients undergoing esophagectomy has also been shown to be associated with an increased incidence of operative mortality and complications [4]. Our study confirms the findings of these previous reports. We noted highly significant independent associations of functional status, increasing patient age, and diabetes mellitus with perioperative morbidity and mortality. When intraoperative variables are included in the analysis, the need for intraoperative transfusion was highly predictive of postoperative mortality and morbidity. This finding also confirms multiple previous reports in the literature [1]. We also noted a significant association between the presence of ascites, elevated alkaline phosphatase concentration, alcohol abuse, and elevated BUN concentration and increased perioperative mortality. The presence of ascites, alcoholism, and elevated alkaline phosphatase concentration, if taken as surrogates of impaired hepatic function, is consistent with findings reported by Bartels and colleagues [15]. Elevated BUN has not been reported previously to be predictive of postoperative mortality. In this study, we noted an independent association of neoadjuvant therapy with perioperative mortality. This association persisted when intraoperative variables were included in the regression analysis. This finding is consistent with other reports in the literature. Six prospective randomized studies of trimodality therapy compared with surgery alone have been reported [18 –23]. Only the study by Walsh and associates [21] demonstrated a significant long-term survival advantage conferred by neoadjuvant chemoradiation therapy. Bosset and coworkers [22], in a study of 297 patients, demonstrated a statistically significant, threefold (12% versus 4%) increase in perioperative mortality in the neoadjuvant chemoradiation therapy group. This increased mortality
Ann Thorac Surg 2003;75:217–22
was thought to be due to an increased incidence of respiratory failure and mediastinal infection in the neoadjuvant therapy group. Two additional studies with smaller numbers of patients in each arm also demonstrated striking differences in perioperative mortality that did not reach statistical significance. Walsh and associates [21] found a 2.8-fold increase in perioperative mortality (10.7% versus 3.7%) in the trimodality group and Nygaard and associates [18] observed a 1.8-fold (24% versus 13%) increase in perioperative mortality in the trimodality group. It is possible that a type II error was present and that had more patients been included in these trials, the differences in perioperative mortality would have reached statistical significance. Alternatively, a recent report by Billingsley and associates [24] retrospectively studied outcomes after trimodality therapy for esophageal carcinoma in the Department of Veterans Affairs Medical Centers. The authors concluded that there was no difference in perioperative mortality between patients who received trimodality therapy and those who received surgery alone. The patients who received neoadjuvant therapy in the Billingsley study, however, were significantly younger, and had significantly fewer comorbidities. The significant differences between the base line characteristics of the patient populations in Billingsley’s study may obscure a real difference in perioperative mortality associated with neoadjuvant therapy. A number of limitations were imposed on this study by the nature of the NSQIP. Whereas the NSQIP reliably and prospectively collects pertinent historical, laboratory, intraoperative, and patient data for surgical procedures covering a number of specialties, the program does not allow for procedure-specific data collection. Some information that would make our analysis more robust such as preoperative variables (pulmonary function tests), operative data (tumor histology and stage), and procedure-specific complications (anastomotic leaks) was not available. We did not specifically study the impact of procedure volume on outcomes. Although the volume of esophagectomies performed at many institutions was low, the procedures at these institutions were often performed by surgeons who perform a high volume of esophagectomies at associated university hospitals. We plan to address this issue, taking into account these factors in a separate report. By doing so we hope to be able to identify the components that contribute to the relationship between case volume and outcome. This series of esophagectomies confirms the feasibility of prospective multicenter investigation of perioperative outcomes. The results, obtained from the Department of Veterans Affairs NSQIP database, largely confirm isolated reports from smaller series in the literature. The -coefficients of individual variables present in individual patients can be used to assess the risk of an adverse perioperative event in that patient. Risk predictions in individual patients, however, should be used in the context of overall clinical judgment. Finally, these find-
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
221
ings can form the foundation for future studies investigating outcomes after esophagectomy.
References 1. Whooley BP, Law S, Murthy SC, et al. Analysis of reduced death and complication rates after esophageal resection. Ann Surg 2001;233:338 –44. 2. Visbal AL, Allen MS, Miller DL, et al. Ivor Lewis esophagogastrectomy for esophageal cancer. Ann Thorac Surg 2001; 71:1803–8. 3. Orringer MB, Marshall B, Iannettoni MD. Transhiatal esophagectomy: clinical experience and refinements. Ann Surg 1999;230:392–403. 4. Karl RC, Schreiber R, Boulware D, et al. Factors affecting morbidity, mortality, and survival in patients undergoing Ivor Lewis esophagogastrectomy. Ann Surg 2000;231:635–43. 5. Hulscher JB, Tijssen JG, Obertop H, van Lanschot JJ. Transthoracic versus transhiatal resection for carcinoma of the esophagus: a meta-analysis. Ann Thorac Surg 2001;72:306 – 13. 6. Altorki N, Skinner D. Should en bloc esophagectomy be the standard of care for esophageal carcinoma? Ann Surg 2001; 234:581–7. 7. Boyle MJ, Franceschi D, Livingstone AS. Transhiatal versus transthoracic esophagectomy: complication and survival rates. Am Surg 1999;65:1137–42. 8. Hagen JA, DeMeester SR, Peters JH, et al. Curative resection for esophageal adenocarcinoma: analysis of 100 en bloc esophagectomies. Ann Surg 2001;234:520 –31. 9. Khuri SF, Daley J, Henderson W, et al. The National Veterans Administration Surgical Risk Study. risk adjustment for the comparative assessment of the quality of surgical care. J Am Coll Surg 1995;180:519 –31. 10. Khuri SF, Daley J, Henderson W, et al. Risk adjustment of the postoperative mortality rate for the comparative assessment of the quality of surgical care: results of the National Veterans Affairs Surgical Risk Study. J Am Coll Surg 1997; 185:315–27. 11. Khuri SF, Daley J, Henderson W, et al. The Department of Veterans Affairs’ NSQIP: the first national, validated, outcome-based, risk-adjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program. Ann Surg 1998;228:491–507. 12. Daley J, Khuri SF, Henderson W, et al. Risk adjustment of the postoperative morbidity rate for the comparative assessment of the quality of surgical care: results of the National Veterans Affairs Surgical Risk Study. J Am Coll Surg 1997; 185:328 –40. 13. Daley J, Forbes MG, Young GJ, et al. Validating risk-adjusted surgical outcomes: site visit assessment of process and structure. National VA Surgical Risk Study. J Am Coll Surg 1997;185:341–51. 14. Agha Z, Lofgren RP, VanRuiswyk JV, Layde PM. Are patients at Veterans Affairs medical centers sicker? A comparative analysis of health status and medical resource use. Arch Intern Med 2000;160:3252–7. 15. Bartels H, Stein HJ, Siewert JR. Preoperative risk analysis and postoperative mortality of oesophagectomy for resectable oesophageal cancer. Br J Surg 1998;85:840 –4. 16. Ferguson MK, Martin TR, Reeder LB, Olak J. Mortality after esophagectomy: risk factor analysis. World J Surg 1997;21: 599 –604. 17. Lund O, Kimose HH, Aagaard MT, et al. Risk stratification and long-term results after surgical treatment of carcinomas of the thoracic esophagus and cardia. A 25-year retrospective study. J Thorac Cardiovasc Surg 1990;99:200 –9. 18. Nygaard K, Hagen S, Hansen HS, et al. Pre-operative radiotherapy prolongs survival in operable esophageal carcinoma: a randomized, multicenter study of pre-operative
222
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
radiotherapy and chemotherapy. The second Scandinavian trial in esophageal cancer. World J Surg 1992;16:1104 –10. 19. Le Prise E, Etienne PL, Meunier B, et al. A randomized study of chemotherapy, radiation therapy, and surgery versus surgery for localized squamous cell carcinoma of the esophagus. Cancer 1994;73:1779 –84. 20. Apinop C, Puttisak P, Preecha N. A prospective study of combined therapy in esophageal cancer. Hepatogastroenterology 1994;41:391–3. 21. Walsh TN, Noonan N, Hollywood D, et al. A comparison of multimodal therapy and surgery for esophageal adenocarcinoma. N Engl J Med 1996;335:462–7.
Ann Thorac Surg 2003;75:217–22
22. Bosset JF, Gignoux M, Triboulet JP, et al. Chemoradiotherapy followed by surgery compared with surgery alone in squamous-cell cancer of the esophagus. N Engl J Med 1997;337:161–7. 23. Urba SG, Orringer MB, Turrisi A, et al. Randomized trial of preoperative chemoradiation versus surgery alone in patients with locoregional esophageal carcinoma. J Clin Oncol 2001;19:305–13. 24. Billingsley KG, Maynard C, Schwartz DL, Dominitz JA. The use of trimodality therapy for the treatment of operable esophageal carcinoma in the veteran population: patient survival and outcome analysis. Cancer 2001;92:1272–80.
DISCUSSION DR DOUGLAS E. WOOD (Seattle, WA): Doctor Bailey, I would like to offer my sincerest congratulations to you and your colleagues for a great presentation and a deserved honor of the J. Maxwell Chamberlain Memorial Paper. The obvious strengths of your study are the independent, validated, and prospective collection of perioperative data and the unbiased multicenter review of all VA esophagectomy patients. Unfortunately, however, all we have learned is that sicker patients have a higher morbidity and mortality. You have added some statistical precision to this knowledge, but I was hoping for more. Perhaps I am overly idealistic, but I had hoped that powerful data such as yours might lead to some suggestions of a risk stratification model for patients or for policy changes within the VA system. Cardiac surgeons are far ahead of general thoracic surgeons in developing risk stratification models to allow a more valid comparison of results. In general, thoracic surgery and the lack of a national database, smaller numbers and more diverse variables have limited our ability to develop similar models. However, in front of us are data that could propel us forward. Do you think you could propose a preliminary risk stratification model for esophagectomy patients based on your data, one that could then be refined as we develop a national general thoracic database? Although I have a great deal of admiration for the VA NSQIP, there are unfortunate limitations in these data. The data points are selected to apply generically to all noncardiac surgical specialties; however, this strategy misses important specialty or procedure-specific data, for example, pulmonary function tests or anastomotic leak. As reported in our meetings last year, mortality after esophagectomy is increased three- to fivefold in low-volume hospitals, which is almost certainly a surrogate for low-volume surgeons. In this series of 1777 patients, esophagectomies were performed in 109 VA hospitals over 10 years, for an average rate of only 1.6 esophagectomies per hospital per year. I would expect that a few centers performed many esophagectomies and that most performed the procedure only rarely; therefore, your mortality and morbidity rates are probably an average of very good outcomes and poor outcomes. Is there an ability in the NSQIP to allow hospital and surgeon experience to be entered as independent variables? Factoring in these variables would yield powerful data indeed, and would be the biggest step toward improving outcomes. In fact, the VA system provides the perfect opportunity to improve results by centralizing specialized procedures where only ego or inconvenience rather than money are the objections to patient referral to a specialized center. Do you
think that your data could be used to influence VA policy for esophagectomy, eventually having these patients referred to esophagectomy centers in the VA system? Doctor Bailey, I like your study and I appreciate the privilege of the discussion. I hope that you will not stop with what you have but take this research to the next step to provide a tool for thoracic surgeons and for policy decisions within and outside the VA. DR BAILEY: Thank you, Dr Wood, for your comments. You asked a number of questions. First with regard to proposing a risk stratification model, we do intend to use the data presented today to proceed in this direction. The report presented here was intended as a starting point to identify risk factors associated with an increased risk of morbidity and mortality after esophageal resection. We plan to create a predicitve, quantitative risk model as well as study a number of other issues in the near future. You asked about procedure-specific data, and that is indeed one of the shortcomings of the protocol for the NSQIP. As it is, the NSQIP is the largest, most well-organized prospective data collection vehicle of which I am aware. With its current limitations, the program is still an enormous undertaking at each VA hospital throughout the country, requiring significant time and money, and small sacrifices need to be made with regard to procedure-specific data culled at each institution. Clearly, if we had more procedure-specific information the database would be more robust; however, we have to accede to some limitations. You asked about volume analysis. Clearly over the last several years, studies using administrative databases have demonstrated an association between better outcomes with highvolume centers. However, this volume– outcome relationship was not observed in a study using the NSQIP looking at eight moderately complicated procedures. The range of esophagectomies per institution in our study was less than one to approximately 14 per year. We have not analyzed number of procedures per institution as an independent variable for our current analysis, but plan to do so in the near future. Finally, you asked about VA policy and establishing esophageal surgery centers as an effort to improve outcomes in the VA medical centers. I think that this idea is reasonable; however, we would like to perform the volume analysis, and demonstrate that a volume– outcome relationship exists before we institute policies mandating specific locations for patients requiring esophageal resection. Thank you.
Outcomes After Esophagectomy: A Ten-Year Prospective Cohort Stephen H. Bailey, MD, David A. Bull, MD, David H. Harpole, MD, Jeffrey J. Rentz, MD, Leigh A. Neumayer, MD, Theodore N. Pappas, MD, Jennifer Daley, MD, William G. Henderson, PhD, Barbara Krasnicka, PhD, and Shukri F. Khuri, MD Veterans Affairs Medical Center, University of Utah Medical School, Salt Lake City, Utah; Veterans Affairs Medical Center, Duke University Medical School, Durham, North Carolina; Institute for Health Policy, Massachusetts General Hospital/Partners Healthcare System and Department of Medicine, Harvard Medical School, Boston, Massachusetts; Veterans Affairs Medical Center, Harvard Medical School, Brockton/West Roxbury, Massachusetts; and Cooperative Studies Program Coordinating Center, Veterans Affairs Medical Center, Hines, Illinois
Background. The Department of Veterans Affairs National Surgical Quality Improvement Program is a unique resource to prospectively analyze surgical outcomes from a cross-section of surgical services nationally. We used this database to assess risk factors for morbidity and mortality after esophagectomy in Veterans Affairs Medical Centers from 1991 to 2001. Methods. A total of 1,777 patients underwent an esophagectomy at 109 Veterans Affairs hospitals with complete in-hospital and 30-day outcomes recorded. Bivariate and multivariable analyses were completed. Results. Thirty-day mortality was 9.8% (174/1,777) and the incidence of one or more of 20 predefined complications was 49.5% (880/1,777). The most frequent postoperative complications were pneumonia in 21% (380/1,777), respiratory failure in 16% (288/1,777), and ventilator support more than 48 hours in 22% (387/1,777). Preoperative predictors of mortality based on multivariable analysis included neoadjuvant therapy, blood urea nitrogen level
of more than 40 mg/dL, alkaline phosphatase level of more than 125 U/L, diabetes mellitus, alcohol abuse, decreased functional status, ascites, and increasing age. Preoperative factors impacting morbidity were increasing age, dyspnea, diabetes mellitus, chronic obstructive pulmonary disease, alkaline phosphatase level of more than 125 U/L, lower serum albumin concentration, increased complexity score, and decreased functional status. Intraoperative risk factors for mortality included the need for transfusion; intraoperative risk factors for morbidity included the need for transfusion and longer operative time. Conclusions. These data constitute the largest prospective outcomes cohort in the literature and document a near 50% morbidity rate and 10% mortality rate after esophagectomy. Data from this study can be used to better stratify patients before esophagectomy. (Ann Thorac Surg 2003;75:217–22) © 2003 by The Society of Thoracic Surgeons
E
models of morbidity and mortality allowing comparative assessments of quality of care among institutions nationwide.
sophageal resection is associated with high rates of perioperative morbidity and mortality. Most series reporting outcomes after esophagectomy are from single institutions and have small numbers of patients with varying comorbidities. These drawbacks make it difficult to extrapolate the results of these series to the general population of patients undergoing esophageal resection nationally [1– 8]. As a result, defining risk factors associated with adverse perioperative outcomes is problematic. The Department of Veterans Affairs (VA) National Surgical Quality Improvement Program (NSQIP) was designed to reliably and prospectively collect pertinent historical, laboratory, intraoperative, and patient data for surgical procedures covering a number of specialties [9 –13]. The goal of the NSQIP is to develop risk-adjusted Presented at the Thirty-eighth Annual Meeting of The Society of Thoracic Surgeons, Fort Lauderdale, FL, Jan 28 –30, 2002. Address reprint requests to Dr Bull, 50 North Medical Drive, University of Utah, Division of Cardiothoracic Surgery, Room 3C127, Salt Lake City, UT 84132; e-mail: [email protected].
© 2003 by The Society of Thoracic Surgeons Published by Elsevier Science Inc
Patients and Methods The methods of the VA NSQIP have been described in detail elsewhere [9 –11].
Population Data were acquired prospectively from 109 Department of Veterans Affairs Medical Centers performing esophagectomy between January 1991 and December 2000. The database included 1,777 patients who had undergone esophageal resection (current procedural terminology codes 43107 to 43124). Esophagectomy was performed for malignancy in 84.9% (n ⫽ 1,509) of cases and benign disease in 15.1% (n ⫽ 268). The number of procedures per institution ranged from 1 to 55 and 33 institutions performed 20 or more resections. 0003-4975/03/$30.00 PII S0003-4975(02)04368-0
218
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Patient Characteristics A dedicated, trained nurse reviewer recorded 122 individual variables for each operation. Each reviewer was trained and continually tested and monitored on a set of standard criteria and definitions for coding all collected variables. These variables were selected by an expert advisory panel and from literature reviews for the NSQIP. Most laboratory values and demographic variables were downloaded directly to the statistical center from computers at each participating VA Medical Center. A series of data audits were performed on an ongoing basis to ensure data completeness and reliability. Each center underwent periodic internal and random external audits for data quality. A more detailed description of the auditing process is available elsewhere [11].
Statistical Analysis Bivariate analyses were performed relating the demographic, laboratory, and preoperative variables to 30-day morbidity and mortality. The unpaired t test was used for continuous variables and the 2 test was used for categorical variables. Variables with a prevalence more than 0.5% and that were significant on bivariate analysis at p ⬍ 0.20 were considered potential independent variables in a multivariable logistic regression. Stepwise logistic regression analysis with entry and exit criteria set at p ⱕ 0.05 was used with mortality as the dependent variable. For morbidity analyses, the presence or absence of at least one complication was considered the dependent variable. The multivariable regression model for 30-day mortality and morbidity was repeated, incorporating intraoperative variables such as operative time and blood loss.
Results Patient Characteristics and Outcomes The study population consisted of 1,777 patients who underwent esophagectomy. The sample was predominantly male (99.1%, n ⫽ 1,761) with a mean age of 63.4 ⫾ 9.9 years. Esophageal resection was performed for a neoplastic process in 84.9% (n ⫽ 1509) and for benign disease in 15.1% (n ⫽ 269). Other patient characteristics are listed in Table 1. Thirty-day mortality was 9.8% (n ⫽ 174) and morbidity was 49.5% (n ⫽ 880). No difference was noted in 30-day mortality when comparing patients with benign disease (10.1%) and patients with malignancy (9.7%). Likewise, 30-day morbidity was not different between patients with benign disease (51.1%) and patients with malignancy (49.0%). The most common complications were pulmonary: pneumonia in 21.4% (n ⫽ 380), ventilator dependence for more than 48 hours in 21.8% (n ⫽ 387), and unplanned reintubation in 16.2% (n ⫽ 288). Other complications are listed in Table 2.
Bivariate Analyses Patient characteristics associated with an increased risk of 30-day mortality (p ⬍ 0.2) included 21 preoperative clinical variables, 10 laboratory values, and 5 periopera-
Ann Thorac Surg 2003;75:217–22
Table 1. Patient Characteristics and Preoperative Risk Factors Variable Male Smoking within 2 weeks ⬎ 10% weight loss Severe COPD More than 2 drinks/day within 2 weeks Dyspnea at rest/minimal exertion Diabetes Partially or totally dependent Disseminated cancer History of CVA with deficit Chronic steroid use Congestive Heart Failure Impaired sensorium Hemiplegia Ascites COPD ⫽ chronic obstructive pulmonary disease; cular accident.
n/N
%
1761/1777 784/1749 569/1777 402/1777 341/1742 341/1759 221/1749 148/1777 142/1777 70/1749 46/1777 35/1777 36/1777 26/1749 10/1749
99.1 44.8 32.0 22.6 19.6 19.4 12.6 8.3 8.0 4.0 2.6 2.0 2.0 1.5 0.6
CVA ⫽ cerebrovas-
tive variables. Fifteen preoperative variables, 12 laboratory values, and 7 perioperative variables were associated with increased risk of 30-day morbidity.
Multivariable Analyses All factors with a prevalence of greater than 0.5% and a bivariate significance of p ⬍ 0.2 were included in multivariable stepwise logistic regression analysis to identify independent risk factors for morbidity and mortality. Preoperative independent predictors of mortality (p ⬍ 0.05) included neoadjuvant therapy, insulin-dependent diabetes mellitus, current alcohol use, decreased functional status, ascites, blood urea nitrogen (BUN) level of more than 40 mg/dL, alkaline phosphatase level of more than 125 U/L, and increasing age. The -coefficients for these variables as well as their p values and odds ratios are found in Table 3. When the model was repeated with the inclusion of intraoperative variables, all of the preoperative factors continued to be independently associated with mortality except for alkaline phosphatase concentration. In this model the need for intraoperative blood transfusion was also predictive of mortality. The results of the multivariable analysis including preoperative and intraoperative factors are shown in Table 4. Each model was then repeated with malignancy added as a separate variable. The presence of malignancy was not associated with an increased risk of 30-day morbidity or mortality (data not shown). Preoperative independent risk factors for overall morbidity (p ⬍ 0.05) on multivariable analysis included increasing age, the presence of dyspnea with mild exertion, insulin-dependent diabetes mellitus, chronic obstructive pulmonary disease (COPD), current smoking, alkaline phosphatase level more than 125 U/L, decreased functional status, lower albumin and higher complexity score. The -coefficient, standard error, p-value and odds ratio are illustrated in Table 5. When this multivariable
Ann Thorac Surg 2003;75:217–22
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Table 2. Thirty-Day Morbidity and Mortality Variable Thirty-day mortality Thirty-day morbidity Wound complications Superficial Deep Dehiscence Respiratory complications Pneumonia Reintubation Pulmonary embolism Failure to wean ⬎ 48 h Renal complications Progressive renal insufficiency Acute renal failure Urinary tract infection CNS complications CVA Coma ⬎ 24 h Peripheral nerve injury Cardiac complications Cardiac arrest Pulmonary edema Myocardial infarction Other complications Prolonged ileus Bleeding requiring ⬎ 4 units PRBC Graft failure DVT Systemic sepsis
n
%
174 880
9.8 49.5
95 100 66
5.3 5.6 3.7
380 288 13 387
21.4 16.2 0.7 21.8
49 37 89
2.8 2.1 5.0
6 11 6
0.3 0.6 0.3
98 64 21
5.5 7.6 1.2
76 98 14 16 169
4.3 5.5 0.8 0.9 9.5
219
Table 4. Multivariable Logistic Regression Model of Preoperative and Intraoperative Variables Predictive of 30Day Mortality Variable
 coefficient
SE
p
OR
⫺5.87 0.05 0.15 2.75 0.43 1.37 0.50 0.47 0.53
0.67 0.01 0.03 0.75 0.13 0.50 0.20 0.18 0.23
0.0001 0.0001 0.0001 0.0002 0.001 0.006 0.01 0.01 0.02
1.05 1.17 15.7 1.53 3.94 1.65 1.60 1.69
Intercept Age Intraop. RBC Ascites Diabetes BUN Alcohol Functional status Neo. therapy ⫺2 Log likelihood ⫽ 1139; statistic ⫽ 8.34 (p ⫽ 0.40).
C-index ⫽ 0.71;
Hosmer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis. BUN ⫽ blood urea nitrogen ⬎ 40 mg/dL; Intraop. RBC ⫽ blood transfusion during surgery; Neo. therapy ⫽ neoadjuvant chemotherapy or radiation therapy; OR ⫽ odds ratio; SE ⫽ standard error.
age, dyspnea, diabetes mellitus, COPD, complexity score, and alkaline phosphatase level of more than 125 U/L (Table 6). The significant intraoperative variables predictive of morbidity included the need for intraoperative blood transfusion, longer operative time, and emergency status.
Model Validation
analysis was repeated incorporating intraoperative variables, preoperative variables that continued to be independently associated with morbidity included increasing
The predictive validity of these models was measured by the c-index. The c-index for the mortality analysis was 0.69 for the model incorporating only preoperative factors and 0.71 when preoperative and intraoperative variables were included. The c-index was 0.62 for the morbidity model when only preoperative variables were included and 0.65 for the model including preoperative and intraoperative variables. These c-indices indicate a moderate level of predictability. The model fit was mea-
Table 3. Multivariable Logistic Regression Model of Preoperative Variables Predictive of 30-Day Mortality
Table 5. Multivariable Logistic Regression Model Including Preoperative Variables Predictive of 30-Day Morbidity
CNS ⫽ central nervous system; CVA ⫽ cerebrovascular accident; DVT ⫽ deep vein thrombosis; PRBC ⫽ packed red blood cells.
Variable
 coefficient
SE
p
OR
⫺5.65 0.05 3.03 0.42 0.48 0.62 1.30 0.48 0.45
0.66 0.01 0.74 0.13 0.18 0.22 0.50 0.20 0.22
0.001 0.0001 0.0001 0.002 0.007 0.01 0.01 0.02 0.05
1.05 20.66 1.52 1.62 1.85 3.67 1.62 1.56
Intercept Age Ascites Diabetes Functional status Neo. therapy BUN Alcohol Alk phos ⫺2 Log likelihood ⫽ 1139; statistic ⫽ 3.01 (p ⫽ 0.93).
C-index ⫽ 0.69;
Hosmer-Lemeshow
Variable
 coefficient
SE
p
OR
⫺1.31 0.38 0.31 0.11 ⫺0.21 0.30 0.36 0.01 0.29
0.34 0.10 0.12 0.04 0.09 0.13 0.15 0.005 0.15
0.0001 0.0001 0.01 0.01 0.02 0.02 0.02 0.03 0.05
1.47 1.36 1.11 0.81 1.36 1.43 1.01 1.34
Intercept Diabetes Dyspnea Complexity score Albumin COPD Alk phos Age Functional status ⫺2 Log likelihood ⫽ 2463; statistic ⫽ 11.20 (p ⫽ 0.19).
C-index ⫽ 0.62;
Homer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis.
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis.
Alk phos ⫽ serum alkaline phosphatase ⬎ 125 U/L; BUN ⫽ blood urea nitrogen ⬎ 40 mg/dL; Neo. therapy ⫽ Neoadjuvant chemotherapy or radiation therapy; OR ⫽ odds ratio; SE ⫽ standard error.
Alk phos ⫽ serum alkaline phosphatase ⬎ 125 U/L; obstructive pulmonary disease; OR ⫽ odds ratio; error.
COPD ⫽ chronic SE ⫽ standard
220
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
Ann Thorac Surg 2003;75:217–22
Table 6. Multivariable Regression Models Including Preoperative and Intraoperative Variables Predictive of 30Day Morbidity Variable
 coefficient
SE
p
OR
⫺1.88 0.15 0.38 0.36 0.30 0.01 0.06 0.10 0.37 0.72
0.38 0.03 0.10 0.13 0.12 0.005 0.02 0.04 0.15 0.31
0.0001 0.0001 0.0002 0.004 0.01 0.02 0.02 0.02 0.02 0.02
1.16 1.46 1.44 1.36 1.01 1.06 1.10 1.44 2.85
Intercept Intraop. PRBC Diabetes COPD Dyspnea Age Operative time Complexity score Alk phos Emergency ⫺2 Log likelihood ⫽ 2463; statistic ⫽ 6.40 (p ⫽ 0.60).
C-index ⫽ 0.65;
Homer-Lemeshow
The dichotomous risk variables were all coded as 0 ⫽ absent, 1 ⫽ present in the logistic regression analysis. Alk phos ⫽ serum alkaline phosphatase ⬎ 125 mg/dL; Intraop. PRBC ⫽ intraoperative packed red blood cells (blood transfusion during surgery); OR ⫽ odds ratio; SE ⫽ standard error.
sured with the Hosmer-Lemeshow statistic. The HosmerLemeshow statistic for the mortality model incorporating preoperative variables was 3.01 (p ⫽ 0.93) and for the mortality model incorporating preoperative and intraoperative variables was 8.34 (p ⫽ 0.40). The HosmerLemeshow statistic was 11.20 (p ⫽ 0.19) for the morbidity model incorporating only preoperative variables and for the morbidity model including preoperative and intraoperative variables was 6.40 (p ⫽ 0.60). These HosmerLemeshow statistics indicate a good level of fit for each model.
Comment In this study, we analyzed the factors affecting perioperative morbidity and mortality in 1,777 patients undergoing esophagectomy in 109 VA Medical Centers during a 10-year period. This report represents the largest prospective study of perioperative outcomes after esophagectomy. The VA NSQIP offers a unique opportunity to study perioperative outcomes because of the large number of patients enrolled and the reliable prospective collection of data by trained and audited nurses. Our analysis demonstrated a 50% perioperative morbidity and 10% perioperative mortality after esophagectomy. These numbers are within the range of recent results in the published literature. High-volume singleinstitution series trend toward the lowest incidence of adverse perioperative events. Orringer and colleagues [3] reported a 4% mortality and approximately 26% morbidity in a series of 1,085 patients undergoing transhiatal esophagectomy. Hagen and associates [8] reported a perioperative mortality of 6% in a series of 100 transthoracic esophagectomies for esophageal adenocarcinoma. Altorki and Skinner [6] demonstrated a 49% morbidity and 5% mortality in a series of 111 patients who underwent transthoracic esophagectomy with radical lymph-
adenectomy for esophageal malignancy. Boyle and colleagues [7] reported a series including both transhiatal and transthoracic esophagectomies with perioperative morbidity of 60% and mortality of 12%. A number of factors contributed to the morbidity and mortality observed in this study. Comorbidity was common among the study patients. The most common comorbidities were active smoking (44%), alcohol abuse (19%), dyspnea at rest or with minimal exertion (19%), severe COPD (22%), chronic steroid use (10%), and disseminated cancer (8%). Data exist to support the hypothesis that patients cared for in VA Medical Centers are indeed more ill at base line than the general population [14]. Further, most prior studies in the literature reflect outcomes after elective operations. By contrast, our series included 54 patients in whom the procedure was performed under emergent circumstances. Relatively few studies in the literature have used multivariable analysis to evaluate risk factors for adverse perioperative outcomes after esophagectomy. These studies have consistently demonstrated that decreased functional status is a significant predictor of poor outcomes after esophagectomy [15–17]. Advanced patient age has been shown to be associated with an increased risk of perioperative morbidity and mortality after esophagectomy [15, 17]. Finally, the presence of diabetes mellitus in patients undergoing esophagectomy has also been shown to be associated with an increased incidence of operative mortality and complications [4]. Our study confirms the findings of these previous reports. We noted highly significant independent associations of functional status, increasing patient age, and diabetes mellitus with perioperative morbidity and mortality. When intraoperative variables are included in the analysis, the need for intraoperative transfusion was highly predictive of postoperative mortality and morbidity. This finding also confirms multiple previous reports in the literature [1]. We also noted a significant association between the presence of ascites, elevated alkaline phosphatase concentration, alcohol abuse, and elevated BUN concentration and increased perioperative mortality. The presence of ascites, alcoholism, and elevated alkaline phosphatase concentration, if taken as surrogates of impaired hepatic function, is consistent with findings reported by Bartels and colleagues [15]. Elevated BUN has not been reported previously to be predictive of postoperative mortality. In this study, we noted an independent association of neoadjuvant therapy with perioperative mortality. This association persisted when intraoperative variables were included in the regression analysis. This finding is consistent with other reports in the literature. Six prospective randomized studies of trimodality therapy compared with surgery alone have been reported [18 –23]. Only the study by Walsh and associates [21] demonstrated a significant long-term survival advantage conferred by neoadjuvant chemoradiation therapy. Bosset and coworkers [22], in a study of 297 patients, demonstrated a statistically significant, threefold (12% versus 4%) increase in perioperative mortality in the neoadjuvant chemoradiation therapy group. This increased mortality
Ann Thorac Surg 2003;75:217–22
was thought to be due to an increased incidence of respiratory failure and mediastinal infection in the neoadjuvant therapy group. Two additional studies with smaller numbers of patients in each arm also demonstrated striking differences in perioperative mortality that did not reach statistical significance. Walsh and associates [21] found a 2.8-fold increase in perioperative mortality (10.7% versus 3.7%) in the trimodality group and Nygaard and associates [18] observed a 1.8-fold (24% versus 13%) increase in perioperative mortality in the trimodality group. It is possible that a type II error was present and that had more patients been included in these trials, the differences in perioperative mortality would have reached statistical significance. Alternatively, a recent report by Billingsley and associates [24] retrospectively studied outcomes after trimodality therapy for esophageal carcinoma in the Department of Veterans Affairs Medical Centers. The authors concluded that there was no difference in perioperative mortality between patients who received trimodality therapy and those who received surgery alone. The patients who received neoadjuvant therapy in the Billingsley study, however, were significantly younger, and had significantly fewer comorbidities. The significant differences between the base line characteristics of the patient populations in Billingsley’s study may obscure a real difference in perioperative mortality associated with neoadjuvant therapy. A number of limitations were imposed on this study by the nature of the NSQIP. Whereas the NSQIP reliably and prospectively collects pertinent historical, laboratory, intraoperative, and patient data for surgical procedures covering a number of specialties, the program does not allow for procedure-specific data collection. Some information that would make our analysis more robust such as preoperative variables (pulmonary function tests), operative data (tumor histology and stage), and procedure-specific complications (anastomotic leaks) was not available. We did not specifically study the impact of procedure volume on outcomes. Although the volume of esophagectomies performed at many institutions was low, the procedures at these institutions were often performed by surgeons who perform a high volume of esophagectomies at associated university hospitals. We plan to address this issue, taking into account these factors in a separate report. By doing so we hope to be able to identify the components that contribute to the relationship between case volume and outcome. This series of esophagectomies confirms the feasibility of prospective multicenter investigation of perioperative outcomes. The results, obtained from the Department of Veterans Affairs NSQIP database, largely confirm isolated reports from smaller series in the literature. The -coefficients of individual variables present in individual patients can be used to assess the risk of an adverse perioperative event in that patient. Risk predictions in individual patients, however, should be used in the context of overall clinical judgment. Finally, these find-
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
221
ings can form the foundation for future studies investigating outcomes after esophagectomy.
References 1. Whooley BP, Law S, Murthy SC, et al. Analysis of reduced death and complication rates after esophageal resection. Ann Surg 2001;233:338 –44. 2. Visbal AL, Allen MS, Miller DL, et al. Ivor Lewis esophagogastrectomy for esophageal cancer. Ann Thorac Surg 2001; 71:1803–8. 3. Orringer MB, Marshall B, Iannettoni MD. Transhiatal esophagectomy: clinical experience and refinements. Ann Surg 1999;230:392–403. 4. Karl RC, Schreiber R, Boulware D, et al. Factors affecting morbidity, mortality, and survival in patients undergoing Ivor Lewis esophagogastrectomy. Ann Surg 2000;231:635–43. 5. Hulscher JB, Tijssen JG, Obertop H, van Lanschot JJ. Transthoracic versus transhiatal resection for carcinoma of the esophagus: a meta-analysis. Ann Thorac Surg 2001;72:306 – 13. 6. Altorki N, Skinner D. Should en bloc esophagectomy be the standard of care for esophageal carcinoma? Ann Surg 2001; 234:581–7. 7. Boyle MJ, Franceschi D, Livingstone AS. Transhiatal versus transthoracic esophagectomy: complication and survival rates. Am Surg 1999;65:1137–42. 8. Hagen JA, DeMeester SR, Peters JH, et al. Curative resection for esophageal adenocarcinoma: analysis of 100 en bloc esophagectomies. Ann Surg 2001;234:520 –31. 9. Khuri SF, Daley J, Henderson W, et al. The National Veterans Administration Surgical Risk Study. risk adjustment for the comparative assessment of the quality of surgical care. J Am Coll Surg 1995;180:519 –31. 10. Khuri SF, Daley J, Henderson W, et al. Risk adjustment of the postoperative mortality rate for the comparative assessment of the quality of surgical care: results of the National Veterans Affairs Surgical Risk Study. J Am Coll Surg 1997; 185:315–27. 11. Khuri SF, Daley J, Henderson W, et al. The Department of Veterans Affairs’ NSQIP: the first national, validated, outcome-based, risk-adjusted, and peer-controlled program for the measurement and enhancement of the quality of surgical care. National VA Surgical Quality Improvement Program. Ann Surg 1998;228:491–507. 12. Daley J, Khuri SF, Henderson W, et al. Risk adjustment of the postoperative morbidity rate for the comparative assessment of the quality of surgical care: results of the National Veterans Affairs Surgical Risk Study. J Am Coll Surg 1997; 185:328 –40. 13. Daley J, Forbes MG, Young GJ, et al. Validating risk-adjusted surgical outcomes: site visit assessment of process and structure. National VA Surgical Risk Study. J Am Coll Surg 1997;185:341–51. 14. Agha Z, Lofgren RP, VanRuiswyk JV, Layde PM. Are patients at Veterans Affairs medical centers sicker? A comparative analysis of health status and medical resource use. Arch Intern Med 2000;160:3252–7. 15. Bartels H, Stein HJ, Siewert JR. Preoperative risk analysis and postoperative mortality of oesophagectomy for resectable oesophageal cancer. Br J Surg 1998;85:840 –4. 16. Ferguson MK, Martin TR, Reeder LB, Olak J. Mortality after esophagectomy: risk factor analysis. World J Surg 1997;21: 599 –604. 17. Lund O, Kimose HH, Aagaard MT, et al. Risk stratification and long-term results after surgical treatment of carcinomas of the thoracic esophagus and cardia. A 25-year retrospective study. J Thorac Cardiovasc Surg 1990;99:200 –9. 18. Nygaard K, Hagen S, Hansen HS, et al. Pre-operative radiotherapy prolongs survival in operable esophageal carcinoma: a randomized, multicenter study of pre-operative
222
BAILEY ET AL OUTCOMES AFTER ESOPHAGECTOMY
radiotherapy and chemotherapy. The second Scandinavian trial in esophageal cancer. World J Surg 1992;16:1104 –10. 19. Le Prise E, Etienne PL, Meunier B, et al. A randomized study of chemotherapy, radiation therapy, and surgery versus surgery for localized squamous cell carcinoma of the esophagus. Cancer 1994;73:1779 –84. 20. Apinop C, Puttisak P, Preecha N. A prospective study of combined therapy in esophageal cancer. Hepatogastroenterology 1994;41:391–3. 21. Walsh TN, Noonan N, Hollywood D, et al. A comparison of multimodal therapy and surgery for esophageal adenocarcinoma. N Engl J Med 1996;335:462–7.
Ann Thorac Surg 2003;75:217–22
22. Bosset JF, Gignoux M, Triboulet JP, et al. Chemoradiotherapy followed by surgery compared with surgery alone in squamous-cell cancer of the esophagus. N Engl J Med 1997;337:161–7. 23. Urba SG, Orringer MB, Turrisi A, et al. Randomized trial of preoperative chemoradiation versus surgery alone in patients with locoregional esophageal carcinoma. J Clin Oncol 2001;19:305–13. 24. Billingsley KG, Maynard C, Schwartz DL, Dominitz JA. The use of trimodality therapy for the treatment of operable esophageal carcinoma in the veteran population: patient survival and outcome analysis. Cancer 2001;92:1272–80.
DISCUSSION DR DOUGLAS E. WOOD (Seattle, WA): Doctor Bailey, I would like to offer my sincerest congratulations to you and your colleagues for a great presentation and a deserved honor of the J. Maxwell Chamberlain Memorial Paper. The obvious strengths of your study are the independent, validated, and prospective collection of perioperative data and the unbiased multicenter review of all VA esophagectomy patients. Unfortunately, however, all we have learned is that sicker patients have a higher morbidity and mortality. You have added some statistical precision to this knowledge, but I was hoping for more. Perhaps I am overly idealistic, but I had hoped that powerful data such as yours might lead to some suggestions of a risk stratification model for patients or for policy changes within the VA system. Cardiac surgeons are far ahead of general thoracic surgeons in developing risk stratification models to allow a more valid comparison of results. In general, thoracic surgery and the lack of a national database, smaller numbers and more diverse variables have limited our ability to develop similar models. However, in front of us are data that could propel us forward. Do you think you could propose a preliminary risk stratification model for esophagectomy patients based on your data, one that could then be refined as we develop a national general thoracic database? Although I have a great deal of admiration for the VA NSQIP, there are unfortunate limitations in these data. The data points are selected to apply generically to all noncardiac surgical specialties; however, this strategy misses important specialty or procedure-specific data, for example, pulmonary function tests or anastomotic leak. As reported in our meetings last year, mortality after esophagectomy is increased three- to fivefold in low-volume hospitals, which is almost certainly a surrogate for low-volume surgeons. In this series of 1777 patients, esophagectomies were performed in 109 VA hospitals over 10 years, for an average rate of only 1.6 esophagectomies per hospital per year. I would expect that a few centers performed many esophagectomies and that most performed the procedure only rarely; therefore, your mortality and morbidity rates are probably an average of very good outcomes and poor outcomes. Is there an ability in the NSQIP to allow hospital and surgeon experience to be entered as independent variables? Factoring in these variables would yield powerful data indeed, and would be the biggest step toward improving outcomes. In fact, the VA system provides the perfect opportunity to improve results by centralizing specialized procedures where only ego or inconvenience rather than money are the objections to patient referral to a specialized center. Do you
think that your data could be used to influence VA policy for esophagectomy, eventually having these patients referred to esophagectomy centers in the VA system? Doctor Bailey, I like your study and I appreciate the privilege of the discussion. I hope that you will not stop with what you have but take this research to the next step to provide a tool for thoracic surgeons and for policy decisions within and outside the VA. DR BAILEY: Thank you, Dr Wood, for your comments. You asked a number of questions. First with regard to proposing a risk stratification model, we do intend to use the data presented today to proceed in this direction. The report presented here was intended as a starting point to identify risk factors associated with an increased risk of morbidity and mortality after esophageal resection. We plan to create a predicitve, quantitative risk model as well as study a number of other issues in the near future. You asked about procedure-specific data, and that is indeed one of the shortcomings of the protocol for the NSQIP. As it is, the NSQIP is the largest, most well-organized prospective data collection vehicle of which I am aware. With its current limitations, the program is still an enormous undertaking at each VA hospital throughout the country, requiring significant time and money, and small sacrifices need to be made with regard to procedure-specific data culled at each institution. Clearly, if we had more procedure-specific information the database would be more robust; however, we have to accede to some limitations. You asked about volume analysis. Clearly over the last several years, studies using administrative databases have demonstrated an association between better outcomes with highvolume centers. However, this volume– outcome relationship was not observed in a study using the NSQIP looking at eight moderately complicated procedures. The range of esophagectomies per institution in our study was less than one to approximately 14 per year. We have not analyzed number of procedures per institution as an independent variable for our current analysis, but plan to do so in the near future. Finally, you asked about VA policy and establishing esophageal surgery centers as an effort to improve outcomes in the VA medical centers. I think that this idea is reasonable; however, we would like to perform the volume analysis, and demonstrate that a volume– outcome relationship exists before we institute policies mandating specific locations for patients requiring esophageal resection. Thank you.