Impact of Total Lymph Node Count on Staging and Survival After Gastrectomy for Gastric Cancer: Data From a Large US-Population Database
http://www.100md.com
《临床肿瘤学》
the Division of Biostatistics, City of Hope Cancer Center, Duarte, CA
Division of Surgical Oncology, Cancer Institute of New Jersey, New Brunswick, NJ
ABSTRACT
BACKGROUND: Prognosis of potentially curable (M0), completely resected gastric cancer is primarily determined by pathologic T and N staging criteria. The optimal regional dissection extent during gastrectomy for gastric adenocarcinoma continues to be debated.
METHODS: A gastric cancer data set was created through structured queries to the Surveillance, Epidemiology, and End Results database (1973 to 1999). Relationships between the number of lymph nodes (LNs) examined and survival were analyzed for the stage subgroups T1/2N0, T1/2N1, T3N0, and T3N1.
RESULTS: In every stage subgroup, overall survival was highly dependent on the number of LNs examined. Multivariate prognostic variables in the T1/2N0M0 subgroup were number of LNs examined, age (for both, P < .0001), race (P = .0004), sex (P = .0006), and tumor size (P = .02). A linear trend for superior survival based on more LNs examined could be confirmed for all four stage subgroups. Baseline model–predicted 5-year survival with only one LN examined was 56% (T1/2N0), 35% (T1/2N1), 29% (T3N0), or 13% (T3N1). For every 10 extra LNs dissected, survival improved by 7.6% (T1/2N0), 5.7% (T1/2N1), 11% (T3N0), or 7% (T3N1). A cut-point analysis yielded the greatest survival difference at 10 LNs examined but continued to detect significantly superior survival differences for cut points at up to 40 LNs, always in favor of more LNs examined.
CONCLUSION: Although the impact of stage migration versus improved regional disease control cannot be separated on basis of the available information, the data provide support in favor of extended lymphadenectomy during potentially curative gastrectomy for gastric cancer.
INTRODUCTION
Accurate identification of the extent of disease is an important diagnostic consideration for therapeutic decision-making of any malignancy. Despite its significant decline in incidence over the past half century, gastric cancer in the United States remains a highly lethal disease, with an estimated overall 5-year survival rate of 22%.1 Prognosis of potentially curable (M0), completely resected gastric cancer is primarily determined by pathologic T and N staging criteria.2-5 Nevertheless, the optimal regional dissection extent during curative-intent gastrectomy for gastric adenocarcinoma continues to be debated.6 D2, or extended lymph node dissection (ELND), may carry the potential to provide more appropriate pathologic staging, better regional disease control, and possibly some survival advantage. All these aspects, however, remain unproven, because prospective randomized trials have failed to confirm a clear disease-control and survival benefit to date7,8; evidence in their favor is mainly limited to nonrandomized or single-institution retrospective clinical series.5,6,9-11 Perhaps for this reason, ELND continues to be performed infrequently within the United States, despite the fact that here the majority of patients undergoing curative-intent gastrectomy for gastric cancer presents in advanced stages with nodal metastatic disease12,13 and despite evidence that the total number of lymph nodes (LNs) removed has direct implications on postgastrectomy survival.14 In 1997, the American Joint Committee on Cancer/International Union Against Cancer gastric cancer staging criteria were changed to accommodate an N category assignment, which now reflects the total number of positive LNs, in a range from 0 to 15 or greater.15 The useful prognostic impact of these criteria has been validated in several large clinical series.3,11,16 However, the optimal number of LNs to be removed and examined to achieve an optimum reliability in stage assignment remains less clear. On the basis of various methodologic approaches using different clinical databases, this recommended number has been suggested to be 10,17 15,11,18 or greater than 25.19 Our hypothesis was that there is an optimal number of LNs to be examined for proper staging and survival prediction. In addition, we assumed that this optimal number should be definable in a larger US general population–derived database. We thus investigated the relationship between LN numbers examined after gastrectomy and survival on the basis of information from the Surveillance, Epidemiology, and End Results (SEER) data set, which is published by the National Cancer Institute.
PATIENTS AND METHODS
A gastric cancer data set was created through structured queries to the publicly available version of the SEER database, covering the years 1973-2000.20 The SEER program combines records of 14 cancer registries across the United States. Stage information for gastric adenocarcinomas was created according to the American Joint Committee on Cancer TNM criteria.15 The definitions of stage subcategory assignment are listed in Table 1 . From a cohort of 65,560 gastric cancer patients, individuals were selected based on completeness of clinicopathologic information, M0 status, with at least one LN examined, and exclusive cancer-directed treatment through operation. Those patients who received any component of radiation treatment or who had an unknown grade were excluded. Patients with incomplete resections, or positive margin status (R1, R2), whenever stated, were excluded as well. The stepwise process of data extraction is depicted in Figure 1. Relationships between the number of LNs examined and survival outcomes were analyzed for stage subcategories of interest, which included primarily the node-negative groups T1/2N0 and T3N0, and subsequently subcategories T1/2N1 and T3N1, to interrogate findings for patients with one to six LNs involved. Overall, analyses were only performed for the stage subgroups T1/2N0, T3N0, T1/2N1, and T3N1. T4 lesions were excluded to avoid data heterogeneity. N2 and N3 categories were excluded for similar reasons; in addition, these groups seem to have been staged more appropriately by definition, and a therapeutic benefit of ELND for the N3 category was expected to be significantly smaller.
The primary outcome parameter of interest was overall survival. Survival time as tabulated by SEER was the time from diagnosis until last contact, date of death, or the date used as a cutoff for the SEER database. In the case of each stage subgroup, univariate, multivariate, cut-point, and model-projection survival analyses were performed. Actuarial survival was calculated with the Kaplan-Meier method,21 and univariate comparison between groups was performed by using the log-rank test.22 A regression model to correlate LN counts with survival was fit based on Kaplan-Meier 5-year survival estimates for each LN-count interval. The independent variable was constructed by using the LN-count interval midpoints. Cox regression served as a multivariate technique using a backward elimination model for all covariates.23 The threshold for keeping a variable in the Cox model under backward elimination was P = .05. Independent variables considered for each stage-subgroup–specific Cox multivariate model were grade, T-stage category (for the T1/T2 combination subgroups), number of LNs examined, number of positive LNs (for N1 subgroups only), race, age, sex, tumor size, year of diagnosis, and tumor location. All calculations were performed by using the SAS 8.2 statistical software package (SAS Institute, Cary, NC). Significance of differences was assumed at P < .05.
RESULTS
Patient Selection Based on Pathologic Stage Information
5,109 patients who fulfilled the initial selection criteria and had complete staging information available, those 3,814 representing the four stage subgroups of interest were extracted (Fig 1), including 1,863 node-negative (N0) patients representing two stage subgroups (T1/2N0 [n = 953] and T3N0 [n = 910]) and 1,951 node-positive (N+) patients representing the remaining stage subgroups of interest (T1/2N1 [n = 373] and T3N1 [n = 1,578]). The median number of LNs examined was 8, with a range from 1 to 89. The median follow-up for this cohort was 30 months (range, 1 to 143); surviving patients had a median follow-up of 58 months (range, 1 to 143). Sixty-two percent of patients were male, and the patients' median age at diagnosis was 71 years (range, 18 to 100 years). The primary tumor location was distributed among whole stomach (12%), distal (32%), middle (29%), and proximal tumors (27%). The median (range) of tumor sizes was 4.3 cm (microscopic to 87.0 cm); patients with tumors listed as greater than 20 cm (n = 8) were excluded from the analysis. The tumor-grade spectrum included well-differentiated (7%), moderately differentiated (32%), poorly differentiated (58%), and undifferentiated tumors (3%).
Number of LNs Examined by Stage Subgroup
Absolute and relative frequencies of LN counts by stage subgroup are represented in Figure 2. For patients with either T1/2N0 or T3N0 gastric cancer–stage assignment, the median LN number was 8 (range, 1 to 75). In both stage subgroups, 22% of patients had 15 or more LNs examined, and 13% had 20 or more LNs analyzed. For the T1/2N1 subgroup, the corresponding numbers were 28% and 15%, respectively, with a median LN count of 10. In patients with T3N1 gastric cancer–stage assignment, the median LN count was 9 (range, 1 to 89); 22% of the patients had 15 or more LNs examined, and 12% had 20 or more LNs analyzed. An analysis of variance showed statistically significant differences between the TN subgroups in the mean number of nodes examined (P = .0013). Specifically, T1/2N1 patients had a mean of 12.2 nodes, which was greater than the number of nodes in the T3N0 and T1/2N0 group (both means, 10.3). The mode of LNs examined was six for the whole group, and three, six, one, and six, for T1/2N0, T1/ 2N1, T3N0, and T3N1 subgroups, respectively. In a linear regression analysis on all patients, there was a statistically significant trend (P < .0001) that showed that one positive node was found for each additional five nodes that were examined.
Univariate Survival Analysis
The overall survival results by stage subgroup and number of LNs examined are depicted in Figures 3 to 5. At each of the cutoff points depicted, namely 10, 15, or 25 LNs examined, overall survival was significantly different in every stage subgroup analyzed, always in favor of the higher LN count. Actuarial 5-year survival rates, based on the number of examined LNs, in increasing interval categories of 10 are listed in Table 2. For the T1/2N0 subgroup, 5-year survival increased from 61% (1 to 10 LNs examined) to 93% (> 40 LNs examined). An equal trend was encountered for the other subgroups, with ranges from 33% to 70% (T1/2N1), 33% to 83% (T3N0), and 14% to 50% (T3N1). The only deviating result was encountered in the T1/2N1 subgroup (with 30 to 39 LNs examined), in which survival was based on four individual patients only. Nevertheless, a linear LN count to survival correlation model provided the best fit in each stage subgroup. This linear trend for superior survival based on number of LNs examined is depicted for all four stage subgroups in Figure 6.
Multivariate Survival Analysis
Multivariate analyses of overall survival were conducted for the entire patient cohort, including all stage subgroups, and then separately for each stage subgroup.
Table 3 shows the entire data set in a combined Cox regression analysis. In this multivariate model, we added N and T categories in addition to all other covariates. Statistically significant survival predictors were the number of LNs examined, age at diagnosis, tumor location, race, T category, N category, sex, and year of diagnosis. Tumor size and grade failed to meet significance at P < .05. Multivariate survival analyses then were performed for each stage subgroup individually. In every subgroup, overall survival was highly dependent on the number of LNs examined. For instance, multivariate prognostic variables in the T1/2N0M0 subgroup were number of LNs examined, age (for both, P < .0001), race (P = .0004), sex (P = .0006), and tumor size (P = .02). In this stage subgroup, differences between T1 and T2 categories were not significant. Detailed risk ratios of this analysis are listed in Table 4. The number of LNs examined remained an independent prognostic variable in the multivariate analyses for all other stage subgroups, at significance levels of P .0001. Tables 5 through 7 list summary Cox model results based on the backward elimination-variable selection for the T1/2N1, T3N0, and T3N1 subgroups, respectively. In each subgroup, the number of LNs examined was not only highly predictive of overall survival outcome but also seemed to be the only consistent prognostic parameter in every subgroup. In addition, when the two N1 subgroups were analyzed, the overall number of LNs examined showed greater prognostic impact than the number of positive LNs (Tables 5 and 7).
Cut-Point Survival Analysis
A cut-point analysis was performed to determine the numeric LN value that determines the greatest actuarial survival difference between resulting groups. We selected the ability to detect differences between groups based on the magnitude of the log-rank test 2 statistic. Results for all relevant cut points and stage subgroups are listed in Table 8. In all four stage subgroups, this cut-point analysis yielded the greatest survival difference at the level of 10 or more LNs examined but continued to detect significantly superior survival differences for cut points at up to 40 LNs, always in favor of the higher number of LNs examined.
Projected Numeric LN Impact on Overall Survival
Based on the statistically assumed linearity as best fit, the effect that LN number has on overall survival was calculated. Baseline model-predicted 5-year survival with only one LN examined was 56% (T1/2N0), 35% (T1/2N1), 29% (T3N0), or 13% (T3N1), as listed in Table 9. For every 10 extra LNs dissected, the calculated overall survival improved by 7.6% (T1/2N0), 5.7% (T1/2N1), 11% (T3N0), or 7% (T3N1).
Stage Migration
To test for the effect of stage migration and the number of LNs examined, we performed linear regressions with TNM stage as the dependent variable and the following variables as independent predictors: total LNs examined, number of positive LNs found, and number of negative LNs examined. The regression result showed a proportional increase in TNM stage as the number of LNs examined increased (P = .052). We calculated the number of negative LNs found as the number of positive LNs subtracted from the number of LNs examined. In this regression, as the number of negative LNs increased, TNM staging decreased (P < .0001).
DISCUSSION
The findings indicate that the greater the number of LNs examined, the better was the resulting postgastrectomy survival of T1-3N0-1 gastric cancer. The effect was observed in all stage subgroups analyzed. In addition, there was no isolated cutoff point for numeric LN analysis after gastrectomy for adenocarcinoma, but the trend toward superior survival outcome could be followed to LN counts greater than 40. Regardless of the underlying mechanisms that influence this survival impact of LN counts, the results call for attention to the total LN number analyzed as an important and powerful qualifier of staging information and survival prediction for gastric cancer in future clinical trials. We also think that the data provide strong support in favor of retrieving and analyzing larger numbers of LNs in the context of potentially curative gastrectomy for gastric cancer. However, before generalizing such a conclusion, the mechanism and implications of this phenomenon deserve additional comment.
Undoubtedly, a dominant mechanism that accounts for apparently superior survival after examination of larger LN numbers is stage migration. This effect is synonymous with "inappropriate understaging" of the disease, when too few LNs have been examined. Our analysis was initially designed to investigate precisely this phenomenon by examining stage subgroups T1/2N0 and T3N0 and querying for appropriate N0 assignment, analogous to a study performed on T3N0 colorectal cancer earlier.24 The different LN-number profile among patients of stage subgroups T1/ 2N1 and T3N1 within this data set indicates that a significant portion of patients classified as N0 has been understaged, based on insufficient LN numbers examined. The fact that the greatest cutoff point for all four subgroups was identified at an LN count of 10 also supports that irrespective of the subgroup, the strongest impact of understaging may be observed with LN numbers smaller than 10. But where does understaging end, and where does a therapeutic benefit of extended lymphadenectomy start? Is superior survival of T3N0 gastric cancer, based on 40 compared to 30 LNs examined, still just because of stage migration? It is not possible for us to answer these questions on the basis of our results reflecting the available information. Other authors have attempted to study stage migration and have seen the greatest stage-associated changes with LN counts of less than 10 or 15, dependent in part on whether perigastric LNs alone or second-echelon LNs were included in the analysis.3,14,17,19,25-27 In a Dutch prospective randomized trial, median LN counts were 17 in the D1 (limited LN dissection) group and 30 after D2 dissection. Although N-stage categories were not reported, LN involvement was seen in 54% versus 56% of patients.7 One may conclude that at least the frequency of detected nodal positivity seems unaffected with median LN counts of 17 or greater, which would corroborate assumptions that improved survival of intermediate-stage gastric cancer after ELND is due to a therapeutic benefit rather than stage migration.19 Although there was no obvious overall survival benefit to ELND in the Dutch trial, which may in part be due to a low-volume setting or confounding factors such as a higher splenopancreatectomy rate with D2 dissection and a resulting higher postoperative mortality, certain subgroups such as that with node positivity of N2 category demonstrated superior survival outcomes after D2 dissection.28 Thus, we suspect that in our analysis, the unexpectedly superior survival in all stage subgroups with more than 20 or 25 LNs available for examination is due at least in part to a therapeutic benefit, likely of better regional small-volume disease control. The magnitude of this effect especially in the T3N0 subgroup, however, must surprise, because the recurrence risk of serosa-positive gastric cancer has been found to be at least 50%.4
There are other aspects in this analysis that call for cautious interpretation. Survival outcome is based on a highly selected subset of patients, primarily because of advanced, incurable disease or incomplete staging information in the majority of patients' data queried. Information on margin status or disease histology may be lacking and thus may not be appropriately considered. Analyses with cutoff points of 30 or 40 LNs are based on even fewer individuals. Naturally, this could lead to inherent biases not necessarily suspected to be found inside a large population-based data set. Are patients from high-volume institutions or surgeons overrepresented when patients with high LN numbers are analyzed? Obviously, high surgeon and hospital volume has been linked to improved postgastrectomy mortality rates, more even than surgeon specialty training or interest.29,30 Certainly, LN counts reflect not only the actual number of LNs removed intraoperatively but also (or especially) the number of LNs identified and properly examined during macroscopic and microscopic pathologic analysis. Is the LN count therefore a surrogate for particularly functional, favorable practice patterns, which may extend into better follow-up care? Do we observe superior survival after larger LN numbers have been examined because healthier patients are able to have more LNs removed and because higher LN counts represent patients with lesser comorbidity, fewer general health risks, or who have undergone a more stringent patient-selection process?31 Furthermore, we have to remain aware of the fact that the findings are only valid for those stage subgroups that were examined and that the conclusions drawn are based on a current standard of histopathologic LN analysis that does not take into account qualitative differences in the assessment of individual LNs. The conclusions therefore may not remain valid for qualitative LN analysis such as sentinel LN-assessment techniques.32,33 Finally, the data analyzed have been collected before a more widespread use of postoperative chemoradiotherapy as adjuvant therapy based on the Intergroup Trial 116.34 Interestingly, in this trial, patients had been stratified on the basis of the number of positive LNs (0, 1 to 3, or > 3) but not the total number of LNs analyzed. The surgical undertreatment of the majority of patients entered into this trial, especially with respect to regional lymphadenectomy, has been documented before.35 Thus, "understaging" in the intergroup trial quite likely was as prevalent as in our data set, both of which represent the care standard as practiced throughout the United States. It remains to be demonstrated whether LN counts have a similar effect on survival after multimodality therapy of gastric cancer as they seem to have had in our study after resection alone.
Although the impact of stage migration versus improved regional disease control cannot be separated based on the available information, the data do provide some support in favor of ELND during potentially curative gastrectomy for gastric cancer. Considering these findings, we would urge surgeons as well as pathologists to routinely use a larger number of LNs, hence to remove and examine as many regional LNs as safely feasible. The findings also support looking at overall LN numbers to stratify for survival risk assessment in trials of adjuvant therapy of gastric cancer. This population-based study therefore corroborates findings from other series in which the total LN number analyzed, or the number of negative LNs (or positive-to-negative LN ratio), have been found to convey prognostic information in terms of postgastrectomy survival.19,36,37 Irrespective of whether increased LN counts affect the quality of stage assignment, regional disease control, or both, proper interpretation of postoperative stage-specific survival hazards and the resulting implications for adjuvant therapy are significantly influenced by the number of LNs that are removed and examined during pathologic evaluation.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Jemal A, Tiwari RC, Murray T, et al: Cancer statistics, 2004. CA Cancer J Clin 54:8-29, 2004
Roder JD, Bottcher K, Siewert JR, et al: Prognostic factors in gastric carcinoma: Results of the German Gastric Carcinoma Study 1992. Cancer 72:2089-2097, 1993
Kodera Y, Yamamura Y, Shimizu Y, et al: The number of metastatic lymph nodes: A promising prognostic determinant for gastric carcinoma in the latest edition of the TNM classification. J Am Coll Surg 187:597-603, 1998
Roukos DH, Lorenz M, Karakostas K, et al: Pathological serosa and node-based classification accurately predicts gastric cancer recurrence risk and outcome, and determines potential and limitation of a Japanese-style extensive surgery for Western patients: A prospective with quality control 10-year follow-up study. Br J Cancer 84:1602-1609, 2001
Schwarz RE, Zagala-Nevarez K: Recurrence patterns after radical gastrectomy for gastric cancer: Prognostic factors and implications for postoperative adjuvant therapy. Ann Surg Oncol 9:394-400, 2002
Kodera Y, Schwarz RE, Nakao A: Extended lymph node dissection in gastric carcinoma: Where do we stand after the Dutch and British randomized trials? J Am Coll Surg 195:855-864, 2002
Bonenkamp JJ, Hermans J, Sasako M, et al: Extended lymph-node dissection for gastric cancer. Dutch Gastric Cancer Group. N Engl J Med 340:908-914, 1999
Cuschieri A, Weeden S, Fielding J, et al: Patient survival after D1 and D2 resections for gastric cancer: Long-term results of the MRC randomized surgical trial. Surgical Cooperative Group. Br J Cancer 79:1522-1530, 1999
Siewert JR, Kestlmeier R, Busch R, et al: Benefits of D2 lymph node dissection for patients with gastric cancer and pN0 and pN1 lymph node metastases. Br J Surg 83:1144-1147, 1996
Sasako M, McCulloch P, Kinoshita T, et al: New method to evaluate the therapeutic value of lymph node dissection for gastric cancer. Br J Surg 82:346-351, 1995
Karpeh MS, Leon L, Klimstra D, et al: Lymph node staging in gastric cancer: Is location more important than number? An analysis of 1,038 patients. Ann Surg 232:362-371, 2000
Wanebo HJ, Kennedy BJ, Chmiel J, et al: Cancer of the stomach: A patient care study by the American College of Surgeons. Ann Surg 218:583-592, 1993
Hundahl SA, Menck HR, Mansour EG, et al: The National Cancer Data Base report on gastric carcinoma. Cancer 80:2333-2341, 1997
Hundahl SA, Phillips JL, Menck HR: The National Cancer Data Base Report on poor survival of U.S. gastric carcinoma patients treated with gastrectomy: Fifth Edition American Joint Committee on Cancer staging, proximal disease, and the "different disease" hypothesis. Cancer 88:921-932, 2000
Fleming ID, Cooper JS, Henson DE, et al: American Joint Committee on Cancer Staging Manual (ed 5). Philadelphia, PA, Lippincott-Raven, 1997
Roder JD, Bottcher K, Busch R, et al: Classification of regional lymph node metastasis from gastric carcinoma. German Gastric Cancer Study Group. Cancer 82:621-631, 1998
Bouvier AM, Haas O, Piard F, et al: How many nodes must be examined to accurately stage gastric carcinomas? Results from a population based study. Cancer 94:2862-2866, 2002
Lee HK, Yang HK, Kim WH, et al: Influence of the number of lymph nodes examined on staging of gastric cancer. Br J Surg 88:1408-1412, 2001
Siewert JR, Bottcher K, Stein HJ, et al: Relevant prognostic factors in gastric cancer: Ten-year results of the German Gastric Cancer Study. Ann Surg 228:449-461, 1998
National Cancer Statistics Branch. Surveillance, Epidemiology and End Results (SEER) Program: Public Use CD-ROM (1973-2000). Bethesda, MD, National Cancer Statistics Branch, 2000
Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
Peto R, Pike MC: Conservatism in the approximation E(0-E)2/E in the log rank test for survival data or tumor incidence data. Biometrics 29:579-584, 1973
Cox DR: Regression models and life tables. J R Stat Soc B 34:187-220, 1972
Cserni G, Vinh-Hung V, Burzykowski T: Is there a minimum number of lymph nodes that should be histologically assessed for a reliable nodal staging of T3N0M0 colorectal carcinomas? J Surg Oncol 81:63-69, 2002
Adachi Y, Suematsu T, Shiraishi N, et al: Perigastric lymph node status as a prognostic indicator in patients with gastric cancer. Br J Surg 85:1281-1284, 1998
de Manzoni G, Verlato G, Roviello F, et al: The new TNM classification of lymph node metastasis minimises stage migration problems in gastric cancer patients. Br J Cancer 87:171-174, 2002
Borie F, Plaisant N, Millat B, et al: Appropriate gastric resection with lymph node dissection for early gastric cancer. Ann Surg Oncol 11:512-517, 2004
Hartgrink HH, Van De Velde CJ, Putter H, et al: Extended lymph node dissection for gastric cancer: Who may benefit? Final results of the randomized Dutch gastric cancer group trial. J Clin Oncol 22:2069-2077, 2004
Hannan EL, Radzyner M, Rubin D, et al: The influence of hospital and surgeon volume on in-hospital mortality for colectomy, gastrectomy, and lung lobectomy in patients with cancer. Surgery 131:6-15, 2002
Callahan MA, Christos PJ, Gold HT, et al: Influence of surgical subspecialty training on in-hospital mortality for gastrectomy and colectomy patients. Ann Surg 238:629-636; discussion 636-639, 2003
Dhar DK, Kubota H, Tachibana M, et al: Body mass index determines the success of lymph node dissection and predicts the outcome of gastric carcinoma patients. Oncology 59:18-23, 2000
Hundley JC, Shen P, Shiver SA, et al: Lymphatic mapping for gastric adenocarcinoma. Am Surg 68:931-935, 2002
Song X, Wang L, Chen W, et al: Lymphatic mapping and sentinel node biopsy in gastric cancer. Am J Surg 187:270-273, 2004
Macdonald JS, Smalley SR, Benedetti J, et al: Chemoradiotherapy after surgery compared with surgery alone for adenocarcinoma of the stomach or gastroesophageal junction. N Engl J Med 345:725-730, 2001
Hundahl SA, Macdonald JS, Benedetti J, et al: Surgical treatment variation in a prospective, randomized trial of chemoradiotherapy in gastric cancer: The effect of undertreatment. Ann Surg Oncol 9:278-286, 2002
Kattan MW, Karpeh MS, Mazumdar M, et al: Postoperative nomogram for disease-specific survival after an R0 resection for gastric carcinoma. J Clin Oncol 21:3647-3650, 2003
Nitti D, Marchet A, Olivieri M, et al: Ratio between metastatic and examined lymph nodes is an independent prognostic factor after D2 resection for gastric cancer: Analysis of a large European monoinstitutional experience. Ann Surg Oncol 10:1077-1085, 2003(David D. Smith, Rebecca R)
Division of Surgical Oncology, Cancer Institute of New Jersey, New Brunswick, NJ
ABSTRACT
BACKGROUND: Prognosis of potentially curable (M0), completely resected gastric cancer is primarily determined by pathologic T and N staging criteria. The optimal regional dissection extent during gastrectomy for gastric adenocarcinoma continues to be debated.
METHODS: A gastric cancer data set was created through structured queries to the Surveillance, Epidemiology, and End Results database (1973 to 1999). Relationships between the number of lymph nodes (LNs) examined and survival were analyzed for the stage subgroups T1/2N0, T1/2N1, T3N0, and T3N1.
RESULTS: In every stage subgroup, overall survival was highly dependent on the number of LNs examined. Multivariate prognostic variables in the T1/2N0M0 subgroup were number of LNs examined, age (for both, P < .0001), race (P = .0004), sex (P = .0006), and tumor size (P = .02). A linear trend for superior survival based on more LNs examined could be confirmed for all four stage subgroups. Baseline model–predicted 5-year survival with only one LN examined was 56% (T1/2N0), 35% (T1/2N1), 29% (T3N0), or 13% (T3N1). For every 10 extra LNs dissected, survival improved by 7.6% (T1/2N0), 5.7% (T1/2N1), 11% (T3N0), or 7% (T3N1). A cut-point analysis yielded the greatest survival difference at 10 LNs examined but continued to detect significantly superior survival differences for cut points at up to 40 LNs, always in favor of more LNs examined.
CONCLUSION: Although the impact of stage migration versus improved regional disease control cannot be separated on basis of the available information, the data provide support in favor of extended lymphadenectomy during potentially curative gastrectomy for gastric cancer.
INTRODUCTION
Accurate identification of the extent of disease is an important diagnostic consideration for therapeutic decision-making of any malignancy. Despite its significant decline in incidence over the past half century, gastric cancer in the United States remains a highly lethal disease, with an estimated overall 5-year survival rate of 22%.1 Prognosis of potentially curable (M0), completely resected gastric cancer is primarily determined by pathologic T and N staging criteria.2-5 Nevertheless, the optimal regional dissection extent during curative-intent gastrectomy for gastric adenocarcinoma continues to be debated.6 D2, or extended lymph node dissection (ELND), may carry the potential to provide more appropriate pathologic staging, better regional disease control, and possibly some survival advantage. All these aspects, however, remain unproven, because prospective randomized trials have failed to confirm a clear disease-control and survival benefit to date7,8; evidence in their favor is mainly limited to nonrandomized or single-institution retrospective clinical series.5,6,9-11 Perhaps for this reason, ELND continues to be performed infrequently within the United States, despite the fact that here the majority of patients undergoing curative-intent gastrectomy for gastric cancer presents in advanced stages with nodal metastatic disease12,13 and despite evidence that the total number of lymph nodes (LNs) removed has direct implications on postgastrectomy survival.14 In 1997, the American Joint Committee on Cancer/International Union Against Cancer gastric cancer staging criteria were changed to accommodate an N category assignment, which now reflects the total number of positive LNs, in a range from 0 to 15 or greater.15 The useful prognostic impact of these criteria has been validated in several large clinical series.3,11,16 However, the optimal number of LNs to be removed and examined to achieve an optimum reliability in stage assignment remains less clear. On the basis of various methodologic approaches using different clinical databases, this recommended number has been suggested to be 10,17 15,11,18 or greater than 25.19 Our hypothesis was that there is an optimal number of LNs to be examined for proper staging and survival prediction. In addition, we assumed that this optimal number should be definable in a larger US general population–derived database. We thus investigated the relationship between LN numbers examined after gastrectomy and survival on the basis of information from the Surveillance, Epidemiology, and End Results (SEER) data set, which is published by the National Cancer Institute.
PATIENTS AND METHODS
A gastric cancer data set was created through structured queries to the publicly available version of the SEER database, covering the years 1973-2000.20 The SEER program combines records of 14 cancer registries across the United States. Stage information for gastric adenocarcinomas was created according to the American Joint Committee on Cancer TNM criteria.15 The definitions of stage subcategory assignment are listed in Table 1 . From a cohort of 65,560 gastric cancer patients, individuals were selected based on completeness of clinicopathologic information, M0 status, with at least one LN examined, and exclusive cancer-directed treatment through operation. Those patients who received any component of radiation treatment or who had an unknown grade were excluded. Patients with incomplete resections, or positive margin status (R1, R2), whenever stated, were excluded as well. The stepwise process of data extraction is depicted in Figure 1. Relationships between the number of LNs examined and survival outcomes were analyzed for stage subcategories of interest, which included primarily the node-negative groups T1/2N0 and T3N0, and subsequently subcategories T1/2N1 and T3N1, to interrogate findings for patients with one to six LNs involved. Overall, analyses were only performed for the stage subgroups T1/2N0, T3N0, T1/2N1, and T3N1. T4 lesions were excluded to avoid data heterogeneity. N2 and N3 categories were excluded for similar reasons; in addition, these groups seem to have been staged more appropriately by definition, and a therapeutic benefit of ELND for the N3 category was expected to be significantly smaller.
The primary outcome parameter of interest was overall survival. Survival time as tabulated by SEER was the time from diagnosis until last contact, date of death, or the date used as a cutoff for the SEER database. In the case of each stage subgroup, univariate, multivariate, cut-point, and model-projection survival analyses were performed. Actuarial survival was calculated with the Kaplan-Meier method,21 and univariate comparison between groups was performed by using the log-rank test.22 A regression model to correlate LN counts with survival was fit based on Kaplan-Meier 5-year survival estimates for each LN-count interval. The independent variable was constructed by using the LN-count interval midpoints. Cox regression served as a multivariate technique using a backward elimination model for all covariates.23 The threshold for keeping a variable in the Cox model under backward elimination was P = .05. Independent variables considered for each stage-subgroup–specific Cox multivariate model were grade, T-stage category (for the T1/T2 combination subgroups), number of LNs examined, number of positive LNs (for N1 subgroups only), race, age, sex, tumor size, year of diagnosis, and tumor location. All calculations were performed by using the SAS 8.2 statistical software package (SAS Institute, Cary, NC). Significance of differences was assumed at P < .05.
RESULTS
Patient Selection Based on Pathologic Stage Information
5,109 patients who fulfilled the initial selection criteria and had complete staging information available, those 3,814 representing the four stage subgroups of interest were extracted (Fig 1), including 1,863 node-negative (N0) patients representing two stage subgroups (T1/2N0 [n = 953] and T3N0 [n = 910]) and 1,951 node-positive (N+) patients representing the remaining stage subgroups of interest (T1/2N1 [n = 373] and T3N1 [n = 1,578]). The median number of LNs examined was 8, with a range from 1 to 89. The median follow-up for this cohort was 30 months (range, 1 to 143); surviving patients had a median follow-up of 58 months (range, 1 to 143). Sixty-two percent of patients were male, and the patients' median age at diagnosis was 71 years (range, 18 to 100 years). The primary tumor location was distributed among whole stomach (12%), distal (32%), middle (29%), and proximal tumors (27%). The median (range) of tumor sizes was 4.3 cm (microscopic to 87.0 cm); patients with tumors listed as greater than 20 cm (n = 8) were excluded from the analysis. The tumor-grade spectrum included well-differentiated (7%), moderately differentiated (32%), poorly differentiated (58%), and undifferentiated tumors (3%).
Number of LNs Examined by Stage Subgroup
Absolute and relative frequencies of LN counts by stage subgroup are represented in Figure 2. For patients with either T1/2N0 or T3N0 gastric cancer–stage assignment, the median LN number was 8 (range, 1 to 75). In both stage subgroups, 22% of patients had 15 or more LNs examined, and 13% had 20 or more LNs analyzed. For the T1/2N1 subgroup, the corresponding numbers were 28% and 15%, respectively, with a median LN count of 10. In patients with T3N1 gastric cancer–stage assignment, the median LN count was 9 (range, 1 to 89); 22% of the patients had 15 or more LNs examined, and 12% had 20 or more LNs analyzed. An analysis of variance showed statistically significant differences between the TN subgroups in the mean number of nodes examined (P = .0013). Specifically, T1/2N1 patients had a mean of 12.2 nodes, which was greater than the number of nodes in the T3N0 and T1/2N0 group (both means, 10.3). The mode of LNs examined was six for the whole group, and three, six, one, and six, for T1/2N0, T1/ 2N1, T3N0, and T3N1 subgroups, respectively. In a linear regression analysis on all patients, there was a statistically significant trend (P < .0001) that showed that one positive node was found for each additional five nodes that were examined.
Univariate Survival Analysis
The overall survival results by stage subgroup and number of LNs examined are depicted in Figures 3 to 5. At each of the cutoff points depicted, namely 10, 15, or 25 LNs examined, overall survival was significantly different in every stage subgroup analyzed, always in favor of the higher LN count. Actuarial 5-year survival rates, based on the number of examined LNs, in increasing interval categories of 10 are listed in Table 2. For the T1/2N0 subgroup, 5-year survival increased from 61% (1 to 10 LNs examined) to 93% (> 40 LNs examined). An equal trend was encountered for the other subgroups, with ranges from 33% to 70% (T1/2N1), 33% to 83% (T3N0), and 14% to 50% (T3N1). The only deviating result was encountered in the T1/2N1 subgroup (with 30 to 39 LNs examined), in which survival was based on four individual patients only. Nevertheless, a linear LN count to survival correlation model provided the best fit in each stage subgroup. This linear trend for superior survival based on number of LNs examined is depicted for all four stage subgroups in Figure 6.
Multivariate Survival Analysis
Multivariate analyses of overall survival were conducted for the entire patient cohort, including all stage subgroups, and then separately for each stage subgroup.
Table 3 shows the entire data set in a combined Cox regression analysis. In this multivariate model, we added N and T categories in addition to all other covariates. Statistically significant survival predictors were the number of LNs examined, age at diagnosis, tumor location, race, T category, N category, sex, and year of diagnosis. Tumor size and grade failed to meet significance at P < .05. Multivariate survival analyses then were performed for each stage subgroup individually. In every subgroup, overall survival was highly dependent on the number of LNs examined. For instance, multivariate prognostic variables in the T1/2N0M0 subgroup were number of LNs examined, age (for both, P < .0001), race (P = .0004), sex (P = .0006), and tumor size (P = .02). In this stage subgroup, differences between T1 and T2 categories were not significant. Detailed risk ratios of this analysis are listed in Table 4. The number of LNs examined remained an independent prognostic variable in the multivariate analyses for all other stage subgroups, at significance levels of P .0001. Tables 5 through 7 list summary Cox model results based on the backward elimination-variable selection for the T1/2N1, T3N0, and T3N1 subgroups, respectively. In each subgroup, the number of LNs examined was not only highly predictive of overall survival outcome but also seemed to be the only consistent prognostic parameter in every subgroup. In addition, when the two N1 subgroups were analyzed, the overall number of LNs examined showed greater prognostic impact than the number of positive LNs (Tables 5 and 7).
Cut-Point Survival Analysis
A cut-point analysis was performed to determine the numeric LN value that determines the greatest actuarial survival difference between resulting groups. We selected the ability to detect differences between groups based on the magnitude of the log-rank test 2 statistic. Results for all relevant cut points and stage subgroups are listed in Table 8. In all four stage subgroups, this cut-point analysis yielded the greatest survival difference at the level of 10 or more LNs examined but continued to detect significantly superior survival differences for cut points at up to 40 LNs, always in favor of the higher number of LNs examined.
Projected Numeric LN Impact on Overall Survival
Based on the statistically assumed linearity as best fit, the effect that LN number has on overall survival was calculated. Baseline model-predicted 5-year survival with only one LN examined was 56% (T1/2N0), 35% (T1/2N1), 29% (T3N0), or 13% (T3N1), as listed in Table 9. For every 10 extra LNs dissected, the calculated overall survival improved by 7.6% (T1/2N0), 5.7% (T1/2N1), 11% (T3N0), or 7% (T3N1).
Stage Migration
To test for the effect of stage migration and the number of LNs examined, we performed linear regressions with TNM stage as the dependent variable and the following variables as independent predictors: total LNs examined, number of positive LNs found, and number of negative LNs examined. The regression result showed a proportional increase in TNM stage as the number of LNs examined increased (P = .052). We calculated the number of negative LNs found as the number of positive LNs subtracted from the number of LNs examined. In this regression, as the number of negative LNs increased, TNM staging decreased (P < .0001).
DISCUSSION
The findings indicate that the greater the number of LNs examined, the better was the resulting postgastrectomy survival of T1-3N0-1 gastric cancer. The effect was observed in all stage subgroups analyzed. In addition, there was no isolated cutoff point for numeric LN analysis after gastrectomy for adenocarcinoma, but the trend toward superior survival outcome could be followed to LN counts greater than 40. Regardless of the underlying mechanisms that influence this survival impact of LN counts, the results call for attention to the total LN number analyzed as an important and powerful qualifier of staging information and survival prediction for gastric cancer in future clinical trials. We also think that the data provide strong support in favor of retrieving and analyzing larger numbers of LNs in the context of potentially curative gastrectomy for gastric cancer. However, before generalizing such a conclusion, the mechanism and implications of this phenomenon deserve additional comment.
Undoubtedly, a dominant mechanism that accounts for apparently superior survival after examination of larger LN numbers is stage migration. This effect is synonymous with "inappropriate understaging" of the disease, when too few LNs have been examined. Our analysis was initially designed to investigate precisely this phenomenon by examining stage subgroups T1/2N0 and T3N0 and querying for appropriate N0 assignment, analogous to a study performed on T3N0 colorectal cancer earlier.24 The different LN-number profile among patients of stage subgroups T1/ 2N1 and T3N1 within this data set indicates that a significant portion of patients classified as N0 has been understaged, based on insufficient LN numbers examined. The fact that the greatest cutoff point for all four subgroups was identified at an LN count of 10 also supports that irrespective of the subgroup, the strongest impact of understaging may be observed with LN numbers smaller than 10. But where does understaging end, and where does a therapeutic benefit of extended lymphadenectomy start? Is superior survival of T3N0 gastric cancer, based on 40 compared to 30 LNs examined, still just because of stage migration? It is not possible for us to answer these questions on the basis of our results reflecting the available information. Other authors have attempted to study stage migration and have seen the greatest stage-associated changes with LN counts of less than 10 or 15, dependent in part on whether perigastric LNs alone or second-echelon LNs were included in the analysis.3,14,17,19,25-27 In a Dutch prospective randomized trial, median LN counts were 17 in the D1 (limited LN dissection) group and 30 after D2 dissection. Although N-stage categories were not reported, LN involvement was seen in 54% versus 56% of patients.7 One may conclude that at least the frequency of detected nodal positivity seems unaffected with median LN counts of 17 or greater, which would corroborate assumptions that improved survival of intermediate-stage gastric cancer after ELND is due to a therapeutic benefit rather than stage migration.19 Although there was no obvious overall survival benefit to ELND in the Dutch trial, which may in part be due to a low-volume setting or confounding factors such as a higher splenopancreatectomy rate with D2 dissection and a resulting higher postoperative mortality, certain subgroups such as that with node positivity of N2 category demonstrated superior survival outcomes after D2 dissection.28 Thus, we suspect that in our analysis, the unexpectedly superior survival in all stage subgroups with more than 20 or 25 LNs available for examination is due at least in part to a therapeutic benefit, likely of better regional small-volume disease control. The magnitude of this effect especially in the T3N0 subgroup, however, must surprise, because the recurrence risk of serosa-positive gastric cancer has been found to be at least 50%.4
There are other aspects in this analysis that call for cautious interpretation. Survival outcome is based on a highly selected subset of patients, primarily because of advanced, incurable disease or incomplete staging information in the majority of patients' data queried. Information on margin status or disease histology may be lacking and thus may not be appropriately considered. Analyses with cutoff points of 30 or 40 LNs are based on even fewer individuals. Naturally, this could lead to inherent biases not necessarily suspected to be found inside a large population-based data set. Are patients from high-volume institutions or surgeons overrepresented when patients with high LN numbers are analyzed? Obviously, high surgeon and hospital volume has been linked to improved postgastrectomy mortality rates, more even than surgeon specialty training or interest.29,30 Certainly, LN counts reflect not only the actual number of LNs removed intraoperatively but also (or especially) the number of LNs identified and properly examined during macroscopic and microscopic pathologic analysis. Is the LN count therefore a surrogate for particularly functional, favorable practice patterns, which may extend into better follow-up care? Do we observe superior survival after larger LN numbers have been examined because healthier patients are able to have more LNs removed and because higher LN counts represent patients with lesser comorbidity, fewer general health risks, or who have undergone a more stringent patient-selection process?31 Furthermore, we have to remain aware of the fact that the findings are only valid for those stage subgroups that were examined and that the conclusions drawn are based on a current standard of histopathologic LN analysis that does not take into account qualitative differences in the assessment of individual LNs. The conclusions therefore may not remain valid for qualitative LN analysis such as sentinel LN-assessment techniques.32,33 Finally, the data analyzed have been collected before a more widespread use of postoperative chemoradiotherapy as adjuvant therapy based on the Intergroup Trial 116.34 Interestingly, in this trial, patients had been stratified on the basis of the number of positive LNs (0, 1 to 3, or > 3) but not the total number of LNs analyzed. The surgical undertreatment of the majority of patients entered into this trial, especially with respect to regional lymphadenectomy, has been documented before.35 Thus, "understaging" in the intergroup trial quite likely was as prevalent as in our data set, both of which represent the care standard as practiced throughout the United States. It remains to be demonstrated whether LN counts have a similar effect on survival after multimodality therapy of gastric cancer as they seem to have had in our study after resection alone.
Although the impact of stage migration versus improved regional disease control cannot be separated based on the available information, the data do provide some support in favor of ELND during potentially curative gastrectomy for gastric cancer. Considering these findings, we would urge surgeons as well as pathologists to routinely use a larger number of LNs, hence to remove and examine as many regional LNs as safely feasible. The findings also support looking at overall LN numbers to stratify for survival risk assessment in trials of adjuvant therapy of gastric cancer. This population-based study therefore corroborates findings from other series in which the total LN number analyzed, or the number of negative LNs (or positive-to-negative LN ratio), have been found to convey prognostic information in terms of postgastrectomy survival.19,36,37 Irrespective of whether increased LN counts affect the quality of stage assignment, regional disease control, or both, proper interpretation of postoperative stage-specific survival hazards and the resulting implications for adjuvant therapy are significantly influenced by the number of LNs that are removed and examined during pathologic evaluation.
Authors' Disclosures of Potential Conflicts of Interest
The authors indicated no potential conflicts of interest.
NOTES
Authors' disclosures of potential conflicts of interest are found at the end of this article.
REFERENCES
Jemal A, Tiwari RC, Murray T, et al: Cancer statistics, 2004. CA Cancer J Clin 54:8-29, 2004
Roder JD, Bottcher K, Siewert JR, et al: Prognostic factors in gastric carcinoma: Results of the German Gastric Carcinoma Study 1992. Cancer 72:2089-2097, 1993
Kodera Y, Yamamura Y, Shimizu Y, et al: The number of metastatic lymph nodes: A promising prognostic determinant for gastric carcinoma in the latest edition of the TNM classification. J Am Coll Surg 187:597-603, 1998
Roukos DH, Lorenz M, Karakostas K, et al: Pathological serosa and node-based classification accurately predicts gastric cancer recurrence risk and outcome, and determines potential and limitation of a Japanese-style extensive surgery for Western patients: A prospective with quality control 10-year follow-up study. Br J Cancer 84:1602-1609, 2001
Schwarz RE, Zagala-Nevarez K: Recurrence patterns after radical gastrectomy for gastric cancer: Prognostic factors and implications for postoperative adjuvant therapy. Ann Surg Oncol 9:394-400, 2002
Kodera Y, Schwarz RE, Nakao A: Extended lymph node dissection in gastric carcinoma: Where do we stand after the Dutch and British randomized trials? J Am Coll Surg 195:855-864, 2002
Bonenkamp JJ, Hermans J, Sasako M, et al: Extended lymph-node dissection for gastric cancer. Dutch Gastric Cancer Group. N Engl J Med 340:908-914, 1999
Cuschieri A, Weeden S, Fielding J, et al: Patient survival after D1 and D2 resections for gastric cancer: Long-term results of the MRC randomized surgical trial. Surgical Cooperative Group. Br J Cancer 79:1522-1530, 1999
Siewert JR, Kestlmeier R, Busch R, et al: Benefits of D2 lymph node dissection for patients with gastric cancer and pN0 and pN1 lymph node metastases. Br J Surg 83:1144-1147, 1996
Sasako M, McCulloch P, Kinoshita T, et al: New method to evaluate the therapeutic value of lymph node dissection for gastric cancer. Br J Surg 82:346-351, 1995
Karpeh MS, Leon L, Klimstra D, et al: Lymph node staging in gastric cancer: Is location more important than number? An analysis of 1,038 patients. Ann Surg 232:362-371, 2000
Wanebo HJ, Kennedy BJ, Chmiel J, et al: Cancer of the stomach: A patient care study by the American College of Surgeons. Ann Surg 218:583-592, 1993
Hundahl SA, Menck HR, Mansour EG, et al: The National Cancer Data Base report on gastric carcinoma. Cancer 80:2333-2341, 1997
Hundahl SA, Phillips JL, Menck HR: The National Cancer Data Base Report on poor survival of U.S. gastric carcinoma patients treated with gastrectomy: Fifth Edition American Joint Committee on Cancer staging, proximal disease, and the "different disease" hypothesis. Cancer 88:921-932, 2000
Fleming ID, Cooper JS, Henson DE, et al: American Joint Committee on Cancer Staging Manual (ed 5). Philadelphia, PA, Lippincott-Raven, 1997
Roder JD, Bottcher K, Busch R, et al: Classification of regional lymph node metastasis from gastric carcinoma. German Gastric Cancer Study Group. Cancer 82:621-631, 1998
Bouvier AM, Haas O, Piard F, et al: How many nodes must be examined to accurately stage gastric carcinomas? Results from a population based study. Cancer 94:2862-2866, 2002
Lee HK, Yang HK, Kim WH, et al: Influence of the number of lymph nodes examined on staging of gastric cancer. Br J Surg 88:1408-1412, 2001
Siewert JR, Bottcher K, Stein HJ, et al: Relevant prognostic factors in gastric cancer: Ten-year results of the German Gastric Cancer Study. Ann Surg 228:449-461, 1998
National Cancer Statistics Branch. Surveillance, Epidemiology and End Results (SEER) Program: Public Use CD-ROM (1973-2000). Bethesda, MD, National Cancer Statistics Branch, 2000
Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53:457-481, 1958
Peto R, Pike MC: Conservatism in the approximation E(0-E)2/E in the log rank test for survival data or tumor incidence data. Biometrics 29:579-584, 1973
Cox DR: Regression models and life tables. J R Stat Soc B 34:187-220, 1972
Cserni G, Vinh-Hung V, Burzykowski T: Is there a minimum number of lymph nodes that should be histologically assessed for a reliable nodal staging of T3N0M0 colorectal carcinomas? J Surg Oncol 81:63-69, 2002
Adachi Y, Suematsu T, Shiraishi N, et al: Perigastric lymph node status as a prognostic indicator in patients with gastric cancer. Br J Surg 85:1281-1284, 1998
de Manzoni G, Verlato G, Roviello F, et al: The new TNM classification of lymph node metastasis minimises stage migration problems in gastric cancer patients. Br J Cancer 87:171-174, 2002
Borie F, Plaisant N, Millat B, et al: Appropriate gastric resection with lymph node dissection for early gastric cancer. Ann Surg Oncol 11:512-517, 2004
Hartgrink HH, Van De Velde CJ, Putter H, et al: Extended lymph node dissection for gastric cancer: Who may benefit? Final results of the randomized Dutch gastric cancer group trial. J Clin Oncol 22:2069-2077, 2004
Hannan EL, Radzyner M, Rubin D, et al: The influence of hospital and surgeon volume on in-hospital mortality for colectomy, gastrectomy, and lung lobectomy in patients with cancer. Surgery 131:6-15, 2002
Callahan MA, Christos PJ, Gold HT, et al: Influence of surgical subspecialty training on in-hospital mortality for gastrectomy and colectomy patients. Ann Surg 238:629-636; discussion 636-639, 2003
Dhar DK, Kubota H, Tachibana M, et al: Body mass index determines the success of lymph node dissection and predicts the outcome of gastric carcinoma patients. Oncology 59:18-23, 2000
Hundley JC, Shen P, Shiver SA, et al: Lymphatic mapping for gastric adenocarcinoma. Am Surg 68:931-935, 2002
Song X, Wang L, Chen W, et al: Lymphatic mapping and sentinel node biopsy in gastric cancer. Am J Surg 187:270-273, 2004
Macdonald JS, Smalley SR, Benedetti J, et al: Chemoradiotherapy after surgery compared with surgery alone for adenocarcinoma of the stomach or gastroesophageal junction. N Engl J Med 345:725-730, 2001
Hundahl SA, Macdonald JS, Benedetti J, et al: Surgical treatment variation in a prospective, randomized trial of chemoradiotherapy in gastric cancer: The effect of undertreatment. Ann Surg Oncol 9:278-286, 2002
Kattan MW, Karpeh MS, Mazumdar M, et al: Postoperative nomogram for disease-specific survival after an R0 resection for gastric carcinoma. J Clin Oncol 21:3647-3650, 2003
Nitti D, Marchet A, Olivieri M, et al: Ratio between metastatic and examined lymph nodes is an independent prognostic factor after D2 resection for gastric cancer: Analysis of a large European monoinstitutional experience. Ann Surg Oncol 10:1077-1085, 2003(David D. Smith, Rebecca R)