Proposed Modification of Nodal Status in AJCC Esophageal Cancer Staging System

Transcription

1 ORIGINAL ARTICLES: Proposed Modification of Nodal Status in AJCC Esophageal Cancer Staging System Wayne Hofstetter, MD, Arlene M. Correa, PhD, Neby Bekele, PhD, Jaffer A. Ajani, MD, Alexandria Phan, MD, Ritsuko R. Komaki, MD, Zhongxing Liao, MD, Dipen Maru, MD, Tsung T. Wu, MD, Reza J. Mehran, MD, David C. Rice, MD, Jack A. Roth, MD, Ara A. Vaporciyan, MD, Garrett L. Walsh, MD, Ashleigh Francis, BS, Shanda Blackmon, MD, and Stephen G. Swisher, MD Departments of Thoracic and Cardiovascular Surgery, Gastrointestinal Medical Oncology, Radiation Oncology, Pathology, and Biostatistics, University of Texas MD Anderson Cancer Center, Houston, Texas Background. The current American Joint Committee on Cancer (AJCC) esophageal cancer staging for nodal status is difficult to interpret and is based solely on lymph node location relative to the primary tumor s esophageal location. Recent reports suggest that the number of lymph nodes involved is also an important factor. We reviewed our esophageal experience to propose an improved nodal staging system. Methods. In all, 1,027 patients with resected esophageal cancer from 1970 to 2005 were reviewed. Lymph nodes stations were assigned according to AJCC criteria. Overall survival was assessed by Kaplan-Meier analysis. The impact of location, number of involved lymph nodes, and use of preoperative chemotherapy or radiation therapy, or both, was assessed. Results. Nonregional nodal involvement (n 17) was associated with decreased survival compared with regional (n 441) or celiac nodal (n 73) involvement (3-year: 0% versus 24% and 23%; p < 0.001). The number of involved lymph nodes was strongly associated with survival (3-year: 0 nodes 63%, 1 to 3 nodes 31%, more than 3 nodes 13%; p < 0.001), and multivariable Cox proportional-hazards analysis suggested that the location and number of involved lymph nodes were independent predictors of survival (p < 0.001). We propose a modified nodal staging system that designates celiac nodes as regional and includes number of involved nodes: pn0, no nodes (3 years 63%, n 496); pn1-regional, 1 to 3 nodes (3 years 32%, n 292); pn2-regional, more than 3 nodes (3 years 14%, n 222); pn3-nonregional node (3 years 0%, n 17 [p < ]). This modified nodal staging system better predicts survival than the current AJCC nodal staging system in which survival for pn1 (3 years 24%) and pm1a (3 years 23%) do not differ (p 0.67). The use of induction before surgical resection did not alter the predictive effect of the new nodal staging system. Conclusions. Modification of the AJCC nodal classification system to incorporate the number of involved lymph nodes with regional and nonregional node location simplifies and better predicts long-term survival than does the current AJCC nodal system. (Ann Thorac Surg 2007;84:365 75) 2007 by The Society of Thoracic Surgeons Carcinoma of the esophagus and gastroesophageal junction is an aggressive disease with a poor prognosis [1]. However, there are groups of patients who benefit from therapy. Esophageal cancer staging systems such as the one devised and revised by the American Joint Committee on Cancer (AJCC) [2] are used to stratify the prognosis of patients and are therefore vital for making decisions regarding various treatment algorithms. Therefore, a staging system that poorly stratifies heterogeneous groups of patients would require revision. As is the case, multiple authors have proposed modifications to the current TNM esophageal staging system Accepted for publication Jan 29, Presented at the Fifty-third Annual Meeting of the Southern Thoracic Surgical Association, Tucson, AZ, Nov 8 11, Address correspondence to Dr Hofstetter, University of Texas MD Anderson Cancer Center, Department of Thoracic and Cardiovascular Surgery, Esophageal Cancer Program, 1515 Holcombe Blvd, Box 445, Houston, TX 77030; whofstetter@mdanderson.org. [3 11]. In particular, the nodal, or N, portion of the current TNM staging system is cumbersome and omits critical prognostic information while simultaneously and perhaps unnecessarily complicating the classification of nodal disease within the M category. The number of lymph nodes involved, a factor shown to be important in other publications [9, 12] is not addressed in the current N staging whereas lymph node involvement and visceral metastasis are found grouped within the M staging system. Because of poor outcomes with surgery alone, an increasing number of patients with locoregionally advanced esophageal cancer are treated with preoperative chemoradiation therapy and surgery [13]. The impact of this treatment paradigm shift on the pathologic esophageal cancer staging system (ptnm) has not yet been clearly defined. Because of pathologic downstaging, the possibility exists that chemoradiation therapy treated patients may have different survivals than patients 2007 by The Society of Thoracic Surgeons /07/$32.00 Published by Elsevier Inc doi: /j.athoracsur

2 366 HOFSTETTER ET AL Ann Thorac Surg PROPOSED MODIFICATION OF NODAL STATUS 2007;84: Fig 1. American Joint Commission on Cancer (AJCC) esophageal staging, nodal designation. Used with the permission of the American Joint Committee on Cancer (AJCC), Chicago, Illinois. The original source for this material is the AJCC Cancer Staging Manual, Sixth Edition (2002) published by Springer, New York (available at: Regional Lymph Node Stations for Staging Esophageal Cancer, From Front (A) and Side (B) 1 Supraclavicular nodes Above suprasternal notch and clavicles 2R Right upper paratracheal nodes Between intersection of caudal margin of innominate artery with trachea and the apex of the lung 2L Left upper paratracheal nodes Between top of aortic arch and apex of the lung 3P Posterior mediastinal nodes Upper paraesophageal nodes, above tracheal bifurcation 4R Right lower paratracheal nodes Between intersection of caudal margin of innominate artery with trachea and cephalic border of azygous vein 4L Left lower paratracheal nodes Between top of aortic arch and carina 5 Aortopulmonary nodes Subaortic and para-aortic nodes lateral to the ligamentum arteriosum 6 Anterior mediastinal nodes Anterior to ascending aorta or innominate artery 7 Subcarinal nodes Caudal to the carina of the trachea 8M Middle paraesophageal lymph From the tracheal bifurcation to the caudal margin of the inferior pulmonary vein nodes 8L Lower paraesophageal lymph nodes From the caudal margin of the inferior pulmonary vein to the esophagogastric junction 9 Pulmonary ligament nodes Within the inferior pulmonary ligament 10R Right tracheobronchial nodes From cephalic border of azygous vein to origin of RUL bronchus 10L Left tracheobronchial nodes Between carina and LUL bronchus 15 Diaphragmatic nodes Lying on the dome of the diaphragm, and adjacent to or behind the crura 16 Paracardial nodes Immediately adjacent to the gastroesophageal junction 17 Left gastric nodes Along the course of the left gastric artery 18 Common hepatic nodes Along the course of the common hepatic artery 19 Splenic nodes Along the course of the splenic artery 20 Celiac nodes At the base of the celiac artery

3 Ann Thorac Surg HOFSTETTER ET AL 2007;84: PROPOSED MODIFICATION OF NODAL STATUS Table 1. Patient Characteristics Demographics (n 1,027) Age (median, range) 61 (34 79) Sex Male 857 (83%) Female 170 (17%) Histology Adenocarcinoma 766 (75%) Squamous cell carcinoma 261 (25%) Location Cervical/upper 54 (5%) Middle 153 (15%) Lower/GEJ 820 (80%) Pathologic stage 0 a 136 (13%) I 127 (12%) IIA 223 (22%) IIB 118 (12%) III 324 (32%) IVA 60 (6%) IVB 39 (4%) Lymph nodes involved (49%) (29%) (22%) a No viable tumor in the esophagus after induction therapy (T0). GEJ gastroesophageal junction. treated with surgery alone, and it remains unclear whether ptnm status will predict survival after chemoradiation therapy. In this manuscript, we sought to assess the impact of number of nodes and nodal location on long-term survival. With those factors in mind, we propose a modified nodal staging system for the AJCC ptnm esophageal cancer staging system. In our model, we assume that a modification to a staging system that simplifies the current system, better predicts long-term outcome, and appropriately predicts long-term survival for patients treated with neoadjuvant therapy represents improvement over the current system. Patients and Methods Patients This study included 1,027 patients with histologically confirmed invasive squamous cell carcinoma or adenocarcinoma of the esophagus and gastroesophageal junction who had a minimum of four resected lymph nodes. There were 595 patients who received neoadjuvant therapy, and 432 were treated with surgery alone. All patients were resected with curative intent at the University of Texas M. D. Anderson Cancer Center (MDACC) between 1970 and Patients who did not survive longer than 30 days were not included in this analysis. Patients who were operated on as salvage cases (recurrent cancer after nonsurgical therapy with pathologic complete response 367 and observation) and patients with disease limited to high-grade dysplasia were also not analyzed in this study. This retrospective review of patient data was approved by the MDACC Institutional Review Board, and the need for individual patient consent was waived. Preoperative Staging and Treatment Pretreatment clinical staging involved available technology at the time of diagnosis. Most recently, patients were screened by computed tomography scans, endoscopic ultrasonography, and positron emission tomography. Patients thought to have locally advanced disease were treated with neoadjuvant therapy. Preoperative chemotherapy included three major chemotherapeutic agents: 5-fluorouracil-based, platinum-based, and Taxol-based. Radiation treatment was given to 45 or 50.4 Gy over a 4- to 6-week period and administered with concurrent chemotherapy. Four to 6 weeks after completion of neoadjuvant therapy, patients underwent surgical esophageal resection. Surgical treatment (with or without induction therapy) included Ivor Lewis (abdominal right thoracic approach), three-field (McKeown-type right thoracic, abdominal-cervical approach), or transhiatal esophagectomy (abdominal-cervical approach). Patients treated with surgery alone received surgery without preoperative chemotherapy or radiotherapy. Lymph Node Station Designation Lymph nodes were assigned a station designation according to the AJCC criteria (Fig 1). Nodal stations 2 through 17 were considered regional (N1) lymph nodes. Lymph node stations 18 and 19 and retroperitoneal nodes such as retrocaval or retroaortic lymph nodes were designated as nonregional (M1b) nodes. For esophageal cancers of the distal esophagus or gastroesophageal junction, nodal station 20 (celiac node) was considered metastatic (M1a) disease. Cases involving disease in the upper esophagus with involvement in the supraclavicular, station 1 lymph nodes were similarly designated as M1a. For cases involving the upper or middle esophagus, nodal station 20 was designated nonregional (M1b) disease. Finally, cases of middle to distal cancers with nodal involvement at station 1 were also considered M1b disease. By definition, and for the purposes of our analysis, cases that were designated as M1a or M1b could be either N0 or N1. Statistical Analysis Cross-tabulation was employed on our raw data to assess the minimum number of sampled nodes required to obtain a consistent percentage of lymph node involvement. Survival probability analyses were performed using the Kaplan-Meier method calculated from the date of surgery to the date of death or most recent follow-up. Operative mortality was excluded from survival analyses to allow determination of long-term outcome rather than short-term morbidity. The prognostic significance of potential parameters was determined by univariate analysis. Cox proportional hazards models were fitted for

4 368 HOFSTETTER ET AL Ann Thorac Surg PROPOSED MODIFICATION OF NODAL STATUS 2007;84: Table 2. Univariate Analysis and Survival of Resected Patients According to Nodal Disease Risk Factor N HR (CI) MS (mos) 3-Year Survival 5-Year Survival p Value Current ptnm N (ref) % 52% N (2.30, 3.21) % 14% M1a (2.04, 3.59) % 19% M1b a (4.22, 11.41) 9.2 0% 0% Number of nodes involved (ref) % 52% (1.81, 2.61) % 20% (3.40, 4.98) % 7% Modified N staging N (ref) % 52% N (1.78, 2.57) % 20% N (3.31, 4.87) 12 14% 7% N (4.35, 11.78) 9.1 0% 0% Modified N staging, surgery only N (ref) % 61% N (2.13, 3.83) % 19% N (4.20, 7.66) % 6% N (5.38, 19.58) 9.1 0% 0% Modified N staging, induction therapy surgery N (ref) % 46% N (1.43, 2.32) 23 33% 21% N (2.45, 4.13) % 8% N (2.37, 12.20) 8.4 0% 0% a M1b by nonregional nodal disease only. CI confidence interval; HR hazard ratio; MS median survival. multivariable analysis. Differences between groups were considered statistically significant if the p values were less than 0.05 in a two-tailed test. The statistical analysis was performed with SPSS Software for Windows (Version ; SPSS, Chicago, Illinois). The data were supplied to a second statistician (author N.B.) who was blinded to the staging systems, current versus modified. Analyses were performed to detect superiority (Appendix). Each of the two staging systems was quantified for effectiveness using the log-rank trend test. To assess if one staging system was more effective at differentiation between stages than the other, a permutation test was employed. gastroesophageal junction. Pathologic downstaging after neoadjuvant therapy resulted in a significant number of patients achieving a ptnm stage 0. Unsuspected M1b disease discovered at the time of surgery resulted in a stage IVb designation in 39 patients; 17 of these cases Results Patient and Survival Characteristics The study population included all patients who had squamous cell carcinoma or adenocarcinoma treated from 1970 to 2005 who underwent esophageal resection at MDACC. Mean potential follow-up for the group was 145 months (range, 7.2 to 442). Demographics of the sample group are depicted in (Table 1). As expected for this group of North American patients, the vast majority were male with adenocarcinoma of the distal esophagus or Fig 2. Kaplan-Meier survival, patients grouped by current American Joint Commission on Cancer (AJCC) nodal staging.

5 Ann Thorac Surg HOFSTETTER ET AL 2007;84: PROPOSED MODIFICATION OF NODAL STATUS 369 Kaplan-Meier survival curves show ordered, distinct curves when analyzing survival by number of nodes involved (Fig 3). Multivariable Cox proportional-hazards analysis was performed including T stage, M stage, and number of lymph nodes involved. All three are independent predictive factors of long-term survival (p 0.001). Owing to issues of colinearity, a separate proportional-hazards model was constructed including nodal location and T stage, showing that both factors are independent predictors of survival (p 0.001). Table 3. Current and Proposed Modified AJCC Esophageal Staging Fig 3. Kaplan-Meier survival, patients grouped by number of involved lymph nodes. were due to nonregional nodal metastasis. The majority of patients (66%, 665 of 1,027) were stage II to III. Most patients (56%, 571 of 1,024) were resected by transthoracic approach, 24% (241 of 1,027) had a transhiatal resection, and 13% (132 of 1,027) had three-field resections. The remaining 7% were minimally invasive (1%) or had missing data unaccounted for. We looked at the significance of lymph node location on long-term survival in our cohort of patients as categorized within the current ptnm staging system. Survival at 3 years for patients with N0, N1, M1a, or M1b (by nonregional nodal involvement) disease was 64%, 24%, 23%, and 0%, respectively. Univariate analysis shows that nonregional lymph node metastasis (n 17) was associated with a significantly decreased overall survival compared with regional (n 441) or celiac (n 73) nodal involvement (p 0.001; Table 2). However, the hazard ratios and confidence intervals of the N1 and M1a groups overlap completely (Table 2). Whereas the Kaplan-Meier survival curves (Fig 2) were distinct for nodal stages N0 and M1b, the N1 and M1a survival curves cross and are virtually interchangeable. This is reflected in the 3- and 5-year overall survivals for the N1 and M1a groups of 24% and 23%, and 14% and 19%, respectively. When the data are analyzed in subgroups according to patients who received surgery only or induction therapy, the graphs are nearly identical to that of the whole cohort, and the overlap between the N1 and M1a groups persists (data not shown). To assess the impact of number of lymph nodes involved on long-term survival, metastasis to any lymph node was separated into the following categories: 0, 1 to 3, and more than 3 involved nodes. These groupings were based on previous publications [4, 5, 7, 8, 11, 12]. We found that the number of lymph nodes involved was strongly associated with long-term survival (p 0.001; Table 2). Three-year survival for 0, 1 to 3, and more than 3 lymph nodes was 63%, 31%, and 13%, respectively. Current American Joint Committee on Cancer staging system (ptnm) Primary tumor (T) Tx Primary tumor cannot be assessed T0 No evidence of primary tumor Tis Carcinoma in situ T1 Tumor invades lamina propria or submucosa T2 Tumor invades muscularis propria T3 Tumor invades adventitia T4 Tumor invades adjacent structures Regional lymph nodes (N) Nx Regional nodes cannot be assessed N0 No regional node metastasis N1 Regional node metastasis Distant metastasis (M) Mx Presence of distant metastasis cannot be assessed M0 No distant metastasis M1a Celiac node metastasis for tumors of the lower esophagus/gej, supraclavicular node metastasis for tumors of the upper esophagus M1b Nonregional nodal metastasis or distant metastasis Revised staging system (ptnm) Primary tumor (T) Tx Primary tumor cannot be assessed T0 No evidence of primary tumor Tis Carcinoma in situ T1 Tumor invades lamina propria or submucosa T2 Tumor invades muscularis propria T3 Tumor invades adventitia T4 Tumor invades adjacent structures Regional lymph nodes (N) Nx Regional nodes cannot be assessed N0 0 regional node metastasis N1 1 3 regional lymph node metastases, including nodes previously labeled as M1a N2 3 regional lymph node metastases, including nodes previously labeled as M1a N3 Nonregional lymph node metastasis Distant metastasis (M) Mx Presence of metastasis cannot be assessed M0 No distant metastasis M1 Distant metastasis GEJ gastroesophageal junction.

6 370 HOFSTETTER ET AL Ann Thorac Surg PROPOSED MODIFICATION OF NODAL STATUS 2007;84: Modified Nodal-Status for ptnm Esophageal Cancer Staging System Based on the above findings, modifications to the current AJCC nodal staging system were derived (Table 3). Patients were stratified into four different nodal groups: pn0, no nodal metastasis; pn1, one to three regional lymph node metastasis, inclusive of previously designated M1a nodes; pn2, more than three regional lymph node metastasis, inclusive of M1a nodes; pn3, nonregional nodal involvement (previously M1b). Univariate and multivariable analysis were performed to assess the impact of the modified staging on the prognosis of long-term survival. Kaplan-Meier survival curves were constructed using this proposed staging modification to the nodal status. With the patients grouped by modified N status, 3-year survival was significantly differentiated (p 0.001) between all of the groups at 63%, 32%, 14%, and 0%, respectively (Table 2). Hazard ratios and confidence intervals were distinct between the groups, and Kaplan-Meier analysis of long-term survival shows ordered, distinct survival curves (Fig 4A). Subgroup analysis of this modified grouping using the surgery-only cohort or induction-therapy cohort yielded similar results (Table 2, Fig 4B, C). Multivariable analysis by Cox proportional hazards model was performed to investigate the contribution of the modified nodal status to long-term survival (Table 4). We included in the model the modified nodal groups and the current pt and pm factors and found that the modified nodal system was a strong independent predictive factor for long-term survival (p 0.001). The pt and pm factors remain significant independent prognostic elements as well. Comparison of Modified to Current Nodal Staging Systems The data were supplied to a statistician blinded to which staging group was the current versus the proposed modifications with the queries of (1) are the staging systems effective (do they maintain an ordered differentiation and effectively differentiate the stage groups); and (2) which system is the more effective. Using log-rank trend test and permutation analysis, both systems were found to be ordered and effective; however, the modified system was found to be significantly better than the existing system (p 0.005; see Appendix). Comment The current AJCC esophageal cancer staging system is based on a T (tumor depth), N (regional nodal statues), and M (nonregional nodes or systemic metastases) staging classification [2]. Although the staging system currently in place is good at predicting long-term survival, as indicated by our 4 Fig 4. (A) Kaplan-Meier survival, patients grouped by proposed modification to the American Joint Commission on Cancer nodal staging. (B) Analysis of surgery only subgroup. (C) Analysis of induction therapy subgroup.

7 Ann Thorac Surg HOFSTETTER ET AL 2007;84: PROPOSED MODIFICATION OF NODAL STATUS Table 4. Multivariable Cox Regression Analysis of Modified Nodal Stage and ptnm Factors Risk Factor Frequency HR (CI) p Value a pt T0 (reference) (ref) Tis (0.45, 1.74) 0.72 T (0.60, 1.25) 0.46 T (0.96, 1.90) 0.09 T (1.61, 2.96) T (1.26, 3.94) pm M0 (reference) (ref) M1a (0.77, 1.40) 0.80 M1b (2.70, 6.49) Modified N stage pn0 (reference) (ref) pn (1.29, 1.92) pn (2.05, 3.18) pn (0.67, 2.41) a p value 0.05 accepted as significant. CI confidence interval; HR hazard ratio. data, modifications to the system results in more ordered differentiation of stages and eliminates some of the complicated features of the N and M portions of the TNM system. Revision of the esophageal staging system is a process of continuous evaluation. More data, changing histology, and different treatment algorithms all affect the validity of an existing system. The current proposal to modification is novel as it expands on previous appeals to include the number of positive lymph nodes and also effectively eliminates the cumbersome M1a classification while including all lymph node disease within the N category, reserving the M category for visceral metastatic disease. Furthermore, it incorporates patients treated with induction therapy and effectively predicts survival in that subgroup of patients. Our data confirm the findings of other authors that the number of involved nodes is an independent predictor of long-term survival in esophageal and gastroesophageal junction tumors. In separate publications, Hagen and associates [9] and Lerut and associates [12] showed that en bloc resection resulted in a significant number of resected lymph nodes and that prognosis significantly declined as the number of involved lymph nodes increases. We modeled the numbered cut-offs for the our proposed modifications to be similar to those used in other publications by Korst and associates [4] and Rice and associates [5], and they are also similar to the Japanese nodal staging system for esophageal cancer [8]. We found that when our cohort of patients were categorized by number of involved lymph nodes that there is a monotone relationship to the designations, the hazard ratios and confidence intervals are distinct, and the Kaplan Meier survival curves are ordered and differentiated. These are criteria that an effective staging system must meet. With these factors applied to modify the 371 existing staging system, we found that our proposed nodal staging system was an independent prognostic factor on long-term survival. Furthermore, comparison of the current nodal staging to the proposed system revealed the modified system to be a statistically significant improvement over the current system. Whether three involved nodes is the correct inflection point is not known. However, we agree that determining the optimal number of positive nodes to stratify the staging system is important and is a task that should be completed using a much greater sample size than is available to us here, as that would include more patients within each staging group. An increasing number of patients with locoregionally advanced esophageal cancer (stage II to IVA) are being treated with chemotherapy or chemoradiotherapy before surgery because of poor long-term outcomes with surgery alone [1]. The impact of pathologic downstaging from this preoperative treatment on the ptnm staging system has not been fully evaluated. This manuscript attempts to address the important question of whether a modified ptnm staging is accurate after preoperative chemoradiation. As our results would suggest, the proposed modified nodal staging system is appropriate for patients treated with induction therapy. Kaplan-Meier analysis shows survival curves to be differentiated and distinct, very similar to those patients treated by surgery alone. When analyzing the cohort as a whole, including surgery-only and induction-therapy patients together, multivariable analysis shows the modified nodal system to be a strong predictor of long-term survival. The proposed modifications eliminate the need for an M1a classification and would designate level 20 nodes as regional for lower esophageal or gastroesophageal junction tumors. Analysis of our data shows similar survival of patients with M1a and N1 disease (24% versus 23% 3-year survival). This finding agrees with previously published data by Hagen and associates [12] who report similar survival for patients with celiac or regional lymph node disease after en-bloc esophagectomy (actuarial 5-year survival 28% versus 37%). Both of these manuscripts, ours and that of Hagen and associates [12], would suggest that intermediate survival can be achieved in patients with M1a disease similar to that of regional lymph node metastasis, and that the number of involved lymph nodes is a stronger predictor of survival than M1a nodal location. However, these findings disagree with those of Rice and associates [5] in which the survival of M1a patients is described as equivalent to M1b, and therefore their recommendation is to eliminate M1a category and reclassify these patients as M1b. That may be explained by difficulty in the classification of an M1a lymph node, especially in the celiac position which represents all of the M1a disease in our data. Malaisrie and associates [14] reported data from our own institution describing an intermediate to poor prognosis for patients identified as having celiac adenopathy by endoscopic ultrasonography. Similar findings were reported by Eloubeidi and associates [15] in another singleinstitution experience with endoscopic ultrasonography

8 372 HOFSTETTER ET AL Ann Thorac Surg PROPOSED MODIFICATION OF NODAL STATUS 2007;84: identified celiac adenopathy that resulted in poor prognosis. Given a lack of consensus on how to define the celiac node, it is possible that the cases that have a celiac node defined by the surgeon could differ slightly but significantly in position and prognosis compared with those whose disease is identified by endoscopic ultrasonography. The conflicting data underscore the need to achieve a consensus on the designation of the celiac lymph node and eliminate the M1a distinction within the staging system. There are several notable weaknesses to this manuscript. The current AJCC system accounts for tumor depth, nodal disease, and metastatic disease but does not account for other tumor features such as length, micrometastasis, molecular markers, or various histologic features. Our proposed modifications pertain to nodal status only, and as is true for any staging system, numerous factors must be analyzed to create an entire staging system. Interactions between other factors such as T and M need to be taken into consideration. Our data set is too small to refine the entire staging system; there simply are not enough patients within each stage subgroup to allow for adequate evaluation. This will require evaluation of the T and M and minor classifications (ie, histology) and the use of a larger data set obtained from multiple collaborating institutions, and we urge participation with cooperations currently under way. The relevant strength of the paper lies within the independent and blinded evaluation of our data by a statistician. Through log-rank trend test and permutation test, the revised staging system was found to have better ordered differentiation and more effective differentiation between stages. In conclusion, modification of the AJCC nodal classification system to incorporate the number of involved lymph nodes with regional and nonregional node location simplifies and better predicts long-term survival than does the current AJCC nodal system. Supported by generous grants from the George Sweeney Esophageal Research Fund and the Carl and Ginnie Edwards Research Fund. References 1. Hofstetter W, Swisher SG, Correa AM, et al. Treatment outcomes of resected esophageal cancer. Ann Surg 2002;236: American Joint Committee on Cancer. AJCC cancer staging manual. 6th ed. Philadelphia: Lippincott-Raven, 2002: Steup WH, De Leyn P, Deneffe G, et al. Tumors of the esophagogastric junction. Long term survival in relation to the pattern of lymph node metastasis and a critical analysis of the accuracy or inaccuracy of ptnm classification. J Thorac Cardiovasc Surg 1996;111: Korst RJ, Rusch VW, Venkatraman E, et al. Proposed revision of the staging classification for esophageal cancer. J Thorac Cardiovasc Surg 1998;115: Rice TW, Blackstone EH, Rybicki LA, et al. Refining esophageal cancer staging. J Thorac Cardiovasc Surg 2003;125: de Manzoni G, Pedrazzani C, Verlato G, et al. Comparison of old and new TNM systems for nodal staging in adenocarcinoma of the gastro-oesophageal junction. Br J Surg 2004;91: Ellis FH, Heatley GJ, Krasna MJ, Williamson WA, Balogh K. Esophagogastrectomy for carcinoma of the esophagus and cardia: a comparison of findings and results after standard resection in three consecutive eight-year intervals with improved staging criteria. J Thorac Cardiovasc Surg 1997;113: Kunisaki C, Akiyama H, Nomura M, et al. Developing an appropriate staging system for esophageal carcinoma. J Am Coll Surg 2005; Lerut T, Nafteux P, Moons J, et al. Three-field lymphadenectomy for carcinoma of the esophagus and gastroesophageal junction in 174 Ro resections: impact on staging, disease-free survival, and outcome. A plea for adaptation of TNM classification in upper-half esophageal carcinoma. Ann Surg 2004;240: Swisher SG, Hofstetter WL, Wu T-T, et al. Proposed revision of the esophageal cancer staging system to accommodate pathologic response (pp. following preoperative chemoradiation (CRT). Ann Surg 2005;241: Roder JD, Busch R, Stein HJ, Fink U, Siewert JR. Ratio of invaded to removed lymph nodes as a predictor of survival in squamous cell carcinoma of the oesophagus. Br J Surg 1994;81: Hagen JA, DeMeester SR, Peters JH, Chandrasoma P, De- Meester TR. Curative resection for esophageal adenocarcinoma. Analysis of 100 en bloc esophagectomies. Ann Surg 2001;234: Chirieac LR, Swisher SG, Ajani JA, et al. Post-therapy pathologic stage predicts survival in patients with esophageal carcinoma receiving preoperative chemoradiation. Cancer 2005;103: Malaisrie SC, Hofstetter WL, Correa AM, et al. Endoscopic ultrasonography-identified celiac adenopathy remains a poor prognostic factor despite preoperative chemoradiotherapy in esophageal adenocarcinoma. J Thorac Cardiovasc Surg 2006;131: Eloubeidi MA, Wallace MB, Hoffman BJ, et al. Predictors of survival for esophageal cancer patients with and without celiac axis lymphadenopathy: impact of staging endosonography. Ann Thorac Surg 2001;72: Vauthey J-N, Lauwers GY, Esnaola NF, et al. Simplified staging for hepatocellular carcinoma. J Clin Oncol 2002;20: Smith DD, Schwarz RR, Schwarz RE. Impact of total lymph node count on staging and survival after gastrectomy for gastric cancer: data from a large US-population database. J Clin Oncol 2005;23: Balch CM, Soong SJ, Gershenwald JE, et al. Prognostic factors analysis of 17,600 melanoma patients: validation of the American Joint Committee on Cancer melanoma staging system. J Clin Oncol 2001;19: Bouvier A-M, Haas O, Piard F, Roignot P, Bonithon-Kopp C, Faivre J. How many nodes must be examined to accurately stage gastric carcinomas? Cancer 2002;94: Lagarde SM, ten Kate FJW, Reitsma JB, Busch ORC, van Lancschot JJB. Prognostic Oncol 2006;24: Efron B, Tibshirani R. An introduction to the bootstrap. London: Chapman & Hall, Klein JP, Moeschberger ML. Survival analysis: techniques for censored and truncated data. Berlin: Springer-Verlag, Cantor A. Extending SAS survival analysis techniques for medical research. Cary, NC: SAS Institute, 1997.

9 Ann Thorac Surg HOFSTETTER ET AL 2007;84: PROPOSED MODIFICATION OF NODAL STATUS Appendix Statistical Comparison of AJCC to Modified N-Staging (Neby Bekele, PhD) 373 Fig A1. Distribution of differences under the null distribution. (Obs. Test Stat. observed test statistic.) The main purpose of this discussion is to provide a formal statistical test comparing two staging systems (the standard AJCC nodal staging system for esophageal cancer and a modified AJCC nodal staging system). Before we can compare these two staging systems, we must first define what makes a staging system effective. We also must define how to quantify this effectiveness, and lastly we must have a way of statistically comparing the effectiveness of the two staging systems. The characteristic that defines the effectiveness of any staging system is its ability to differentiate between low-stage patients (those patients who survive a long time), mid-stage patients (patients who moderate amount of time, and high-stage patients (patients who survive a relatively short amount of time). This differentiation is quantified graphically by a monotone relationship between stage of disease and area-under-the survival curve. That is, a good staging system will result in patients categorized has having low-stage disease to also have the most area under the survival curve; patients categorized as having mid-stage disease will have less area under the survival curve compared with the low-stage patients; and lastly, patients categorized as having high-stage disease will have the least amount of area-under-the-survival-curve relative to the other two groups. Moreover, this type of monotone relationship between stage and area-under-the-survival-curve may be quantified numerically by a log-rank trend test [1]. This statistic is used to because it quantifies the effectiveness of a staging system: The more effective a staging system, the larger will be the value of the log-rank trend test. Any test statistic that increases only as (ordered) differentiation between stages increases (as the logrank trend test does) would be adequate for our purposes. As the log-rank trend test is available and can easily be calculated [2], we chose this statistic. Other test statistics such as the standard log-rank test would not be appropriate because this test statistic increases for nonordered differences in survival curves (which for assessing effectiveness of a staging system would not be appropriate). Once we have quantified the effectiveness of each of the two staging systems using the log-rank trend test, we need to also assess if one staging system is more effective at differentiating between stages than the other. The complicating factor in assessing the difference in effectiveness of the two staging systems is that the same set of patients are categorized under the both systems, inducing correlation between the two log-rank trend statistics. We address this complicating factor by assessing differences in the effectiveness of the two staging systems by a permutation (randomization) test. A permutation test is a type of hypothesis test in which the null distribution is obtained by calculating possible values of the test statistic under random rearrangements (ie, permutations) of the original labels assigned to the observed data [3]. By repeating this process many times (eg, 10,000 times), a null distribution is created. Creation of null distributions in this way only differ from null distributions derived from statistical theory (eg, standard normal, 2,F)in how the null distribution is obtained but do not differ in how they are used or interpreted. An added benefit of using null distributions derived from permutation tests is that they can be used in situations in which the null distribution is difficult to construct analytically (as in this case). Each of the staging systems under consideration in this analysis has four staging categories. We call these four categories N0, N1, N2, and N3 in the modified staging system and N0, N1, M1a, and M1b in the current AJCC system. Under the null hypothesis for our permutation test, we assume that the two staging systems are exchangeable. This means that under the null hypothesis we assume that there is no difference in the two staging systems. The alternative hypothesis is that there is difference in the ability of the two staging systems to differentiate among these four stages. To assess these hypotheses, we construct our test statistic, which is the difference in the two trend tests calculated under each staging system. For the observed data, the difference in the two staging system log-rank trend test statistics is ( for the standard AJCC staging system and for the modified AJCC staging system). These log-rank trend tests tell us that both methods are effective staging systems, but it appears that the modified AJCC staging system is better. To assess whether the modified AJCC staging system is in fact statistically better, we construct our null distribution for which this observed test statistic will be compared by performing the following steps: (1) For each patient with 50% probability, randomly rearrange (ie, permute) the staging system labels originally assigned to that patient. That is, for a given patient, the stage assigned under the standard staging system is switched, and the stage assignment is now considered to have been assigned under the modified system and vice versa. (2) Once all patients have been permuted, we calculate the log-rank trend test statistic for the two staging systems and record the permuted difference in log-rank trend tests. (3) We repeat steps one and two 10,000 times. The null distribution we construct using the method described above is given in Appendix Figure 1. As shown, the differences in the trend statistics under the null distribution is centered around 0, as one would expect if there is no difference between staging systems in their ability to differentiate between low-, middle-, and high-stage patients. Moreover, the probability of observing a differences in trend statistics as rare as is only under the null hypothesis. Appendix References 1. Klein JP, Moeschberger ML. Survival analysis: techniques for censored and truncated data. Berlin: Springer-Verlag, Cantor A. Extending SAS survival analysis techniques for medical research. Cary, NC: SAS Institute, Efron B, Tibshirani R. An introduction to the bootstrap. London: Chapman & Hall, 1993.

10 374 HOFSTETTER ET AL Ann Thorac Surg PROPOSED MODIFICATION OF NODAL STATUS 2007;84: DISCUSSION DR PAUL H. SCHIPPER (Portland, OR): Recently there has been an upswelling of discontent with the esophageal cancer staging system, both in T and N staging, and I congratulate Dr Hofstetter and his colleagues for tackling this, and it is a large body of work and thank them for doing that. In light of your data, I would say that N staging is at least an important prognostic indicator and more so than just the presence or absence of disease in an old bed. More importantly, maybe how many nodes are involved? So I have two questions for you. First question. What do you feel is an adequate lymphadenectomy? Should we be removing everything between certain anatomic borders? Should we be aiming for a certain number of nodes removed? What should that number be? Should we be dissecting out the celiac nodes? Second question. Is there a chance that patients with a lesser T stage in your study had fewer nodes resected because of the length of time of this study and because it was retrospective, and because of fewer nodes being resected, that some nodes that were positive were not identified? In other words, how consistent do you think your denominator of lymph nodes resected is? DR HOFSTETTER: I would like to thank Dr Schipper for taking the time to review our paper. Contributions from our reviewers and questions from the audience are appreciated and will strengthen our publication. Regarding the first question, what is an adequate lymph node dissection, our data really does not address that question directly and we did not statistically look at what would be an adequate lymph node dissection. We did set a cut-off for the number of lymph nodes resected in the patients that we looked at. In other words, we wouldn t use patients in this analysis who had less than four lymph nodes evaluated. The cut was set at four because when we looked at our data, we found that patients who had very few lymph nodes resected had a very low incidence of lymph node involvement, perhaps inappropriately low. At four or greater resected lymph nodes, the incidence of finding positive nodes remained consistent; meaning that there may be a minimum number of nodes that need to be resected to decrease a false-negative result, and if you remove less than, say, three, four lymph nodes that you are not going to find lymph node involvement that is there. There have been other publications that allude to a minimum number of lymph nodes involved. In our situation, over 60% of our patients had transthoracic dissections, and our median number of reported lymph nodes in the specimen was 10. So I would say that the extent of the lymph node dissection is important. If you get to the point where you are not dissecting lymph nodes out, you are not going to detect whether the patient has lymph node involvement. Should we be dissecting the celiac lymph node out? I think absolutely. My lymph node dissection comes down through the celiac axis. I clean out the area through the porta hepatis, and extend this left-lateral to the hilum of the spleen. Do I know that that is going to make a difference in long-term survival? I do not, but I believe that it makes a difference in prognosis and there is data to suggest that patients that have limited lymph node involvement do benefit from an extended lymphadenectomy. The second question, does a lesser T stage portend a different prognosis and does that mean that we were doing a lesser operation in terms of a lymphadenectomy? Our practice at this point is that any patient we see who has local regionally advanced disease undergoes an extensive lymph node dissection, and that is independent of T stage. Evidence to that is that we know that patients who have T1b disease in a modified staging system have upwards of 35 to 50% of nodal involvement. Therefore, we are still pushing for an extended lymph node dissection. The only situation where a formal lymph node dissection might be avoided is in a patient with high-grade dysplasia or intramucosal carcinoma. However, for the purposes of this study, we limited our data-set to patients who had lymph nodes removed. DR MARK J. KRASNA (Towson, MD): I want to congratulate Wayne and the group from Anderson on an excellent review. One quick historical comment, Wayne. This idea was proposed first by Dave Skinner in 1988 with his so-called WNM classification. My mentor, Bunkey Ellis, then rekindled this in a paper that we wrote in 1990 that suggested that lymph node prognostication was the most important independent prognosticator and that using the number of nodes should be incorporated. I applaud you because it wasn t included in that last AJCC and hopefully we will get them to change it the next time. I will just ask one specific question. Although I enjoyed your analysis, I would suggest that another separate analysis, and perhaps it was done in the manuscript, be done excluding all the patients who received neoadjuvant therapy. Although your numbers are very high, therefore it is an excellent database to analyze, you actually may be underestimating the degree of nodal extent and therefore underestimating the impact on survival since your neoadjuvant patients may have appeared to have fewer nodes involved at the time of resection. So I wonder if you would comment on whether you have analyzed the overall group without the neoadjuvant patients. DR HOFSTETTER: Thanks, Dr Krasna, for your comments and questions. First of all, thank you for reviewing the historical data, which I didn t have time for in our presentation. Nonetheless, we all recognize your important contributions to the field and appreciate your comments. The concept of including the number of nodes involved into the staging system is clearly not new. It just hasn t been picked up in esophageal staging, even in the most recent AJCC revision in Many of us are interested in seeing this incorporated into a new revision. We did perform a subgroup analysis including only the patients who did not receive neoadjuvant therapy. The Kaplan- Meier survival curves of the patients who underwent surgery only versus induction therapy are included in the manuscript and these show that the relationship of nodal involvement to survival was consistent, whether or not the patient had surgery alone or neoadjuvant therapy. We believe that these proposals for modification of the nodal staging system would include both induction patients and surgery-alone patients. Now, this data doesn t specifically address the importance of downstaging in neoadjuvant therapy patients, but we do think that it is appropriate that downstaged patients be categorized in the same staging system as patients who have undergone surgery alone. Separate data from our group has shown that patients who are downstaged perform to the level of their post-treatment stage, and this manuscript lends some support to this as well. However, this is not the focus of our manuscript. So to answer your question specifically, yes, we did evaluate it in surgery-only patients and the relationship seems to be consistent.

11 Ann Thorac Surg HOFSTETTER ET AL 2007;84: PROPOSED MODIFICATION OF NODAL STATUS DR CAROLYN E. REED (Charleston, SC): Wayne, a great presentation again from the MD Anderson group. I am dissatisfied, as many of us are, with the present staging system, but we have to be careful that the staging system needs to be used clinically as well as pathologically, because we need to talk to our patients clinically ahead of time, we need to use the staging system to pick the appropriate therapy, as well as to talk about the prognosis. So my question is really how do you use what you found, what you have discovered clinically so that we can incorporate it into a new staging system? How is this going to change how you would look at a patient preoperatively rather than after the fact? Thank you. DR HOFSTETTER: Dr Reed, thank you for your comments and questions. Specifically we did deal with pathologic staging because we wanted to get a start on revising the nodal staging system and the overall staging system. This is just a first step in revision of the system. I agree that the staging system should offer prognostic information that is available to clinicians to discuss with their patients prior to making treatment decisions. Ultimately as the entire staging system comes under review, we would like to be able to take that information with us into the clinics. In the manuscript, we discuss the importance of EUS in clinical staging of these patients and how we go on to use this information in terms of offering neoadjuvant therapy. Patients who appear clinically to have minimal to no lymph node involvement may be candidates for a less aggressive approach to induction therapy. However, based on the prognoses seen in this modified staging system, patients who on clinical staging have greater than three lymph nodes involved may be more appropriately treated with an aggressive approach to induction therapy. 375 DR MARK B. ORRINGER (Ann Arbor, MI): My compliments for a nice paper that highlights some of the problems with our current system. Wayne, I would like to ask you if you don t think that part of the reason that your M1a and M1b survival groups overlap is that all M1a patients aren t the same? And isn t there a difference when an M1a, a celiac node, is involved right at the origin of the left gastric artery from the celiac trunk and can be divided flush with the celiac axis and resected en bloc with a specimen versus an independent metastasis, still a celiac node, but going up the portal system heading toward the liver along the hepatic? And so I would ask you if there is not a need to further subdivide what a celiac metastasis is, because that is where I think the big shortcoming of the current system is. All celiac metastases are not the same. DR HOFSTETTER: Dr Orringer, thank you for that comment. I think you have hit on the crux of the discussion for M1a disease, and that is the distinction of how a celiac lymph node is distinguished by the surgeon and other clinicians involved in the work-up of esophageal cancer patients. To my knowledge, there is no standard agreement within the staging system regarding the definition as to what a celiac lymph node is. In fact, from data that we have previously presented, EUS detection of celiac disease portends an intermediate to poor prognosis, and this may be different from outcomes seen when the celiac disease was defined by the surgeon. The lymph node that the surgeon calls a celiac lymph node may be different than that described by the endoscopist. So there is a need to really decipher and pin down what a celiac lymph node is. Perhaps with our collaborative group we can address this question.