Fax: (425) 928-4880
Validation of the 6th edition AJCC pancreatic cancer staging system
Report from the National Cancer Database
Article first published online: 19 JUN 2007
Copyright © 2007 American Cancer Society
Volume 110, Issue 4, pages 738–744, 15 August 2007
How to Cite
Bilimoria, K. Y., Bentrem, D. J., Ko, C. Y., Ritchey, J., Stewart, A. K., Winchester, D. P. and Talamonti, M. S. (2007), Validation of the 6th edition AJCC pancreatic cancer staging system. Cancer, 110: 738–744. doi: 10.1002/cncr.22852
- Issue published online: 2 AUG 2007
- Article first published online: 19 JUN 2007
- Manuscript Accepted: 9 APR 2007
- Manuscript Revised: 29 MAR 2007
- Manuscript Received: 26 FEB 2007
- pancreatic neoplasm;
- National Cancer Data Base
With the development of stage-specific treatments for pancreatic cancer, controversies exist concerning optimal clinical and pathologic staging. The most recent edition of the American Joint Committee on Cancer (AJCC) Cancer Staging Manual 6th Edition included some notable modifications. In anticipation of the 7th edition's publication, the authors evaluated the predictive ability of the current pancreatic adenocarcinoma staging system.
By using the National Cancer Data Base (1992–1998), 121,713 patients were identified with pancreatic adenocarcinoma. All patients were restaged by AJCC 6th edition guidelines. Stage-specific overall survival was estimated by using the Kaplan-Meier method and compared with log-rank tests. Concordance indices were calculated to evaluate the discriminatory power of the staging system. Cox modeling was used to determine the relative impact of T, N, and M classification on survival.
For all patients, there was 5-year survival discrimination by stage (P < .0001). For patients who underwent pancreatectomy, stage predicted 5-year survival: stage IA, 31.4%; IB, 27.2%; IIA, 15.7%; IIB, 7.7%; III, 6.8%; IV, 2.8% (P < .0001). The concordance index for the staging system was 0.631 for all patients, 0.613 for those who underwent pancreatectomy, and 0.596 for patients who did not undergo resection. In patients who underwent pancreatectomy, tumor size, nodal status, and distant metastases were independent predictors of survival (P < .0001).
This is the first large-scale validation of the pancreatic cancer staging system. AJCC 6th edition staging guidelines are accurate with respect to survival. Further investigation is needed to integrate new molecular and biochemical markers into the staging scheme. Cancer 2007; 110:738–44. © 2007 American Cancer Society.
Pancreatic cancer is the 4th leading cause of cancer deaths and the 2nd most common gastrointestinal malignancy in the United States. In 2007, the American Cancer Society estimates that over 37,000 patients will be diagnosed with pancreatic cancer, and 33,000 will die of the disease.1
As less than 20% of pancreatic cancer patients are surgical candidates at presentation, staging systems for pancreatic cancer must be adequate for patients who are staged clinically (nonsurgically based on clinical exam, radiologic evaluation, and laboratory studies) and pathologically (surgically). The American Joint Committee on Cancer (AJCC) has been developing staging guidelines for common solid organ tumors since 1977.2 The most recent AJCC cancer staging manual, the 6th edition, was published in 2002. Development of the 7th edition is underway with an expected publication date in early 2009 and an implementation date for national cancer registries of January 1, 2010.
Numerous neoplasms in the 6th edition staging manual contain empirical verification of site-specific staging guidelines. No such data have been put forth for pancreatic cancer from cancer registries, and the current recommendations are based on large, single-institution studies. From the 5th to the 6th edition, there were 3 notable modifications of the pancreatic cancer staging system: T1 NO MO and T2 NO MO were divided into stages IA and IB, respectively; T1-3 N1 M0 disease moved from stage III to stage IIB; and T4 Nx M0 (x = any) moved from stage IVA to stage III. Moreover, a change in T-classification definitions was made to emphasize the distinction between potentially resectable (T3) and locally advanced (T4) disease. As the 6th edition staging manual has only been in use since 2003, 5-year follow-up information is not available to evaluate survival for patients whowere staged at diagnosis based on 6th edition guidelines.
The National Cancer Data Base (NCDB) is a program of the American College of Surgeons and receives operational support from the American Cancer Society. The NCDB has been collecting data on newly diagnosed cancers since 1985, and now it contains information on more than 20 million patients from more than 1440 hospitals. The NCDB collects information on patient demographics, diagnosis, treatment, staging, recurrence, survival, and health systems information. On the basis of national incidence estimates from the American Cancer Society, the NCDB captures 76% of all new pancreatic cancer diagnoses in the United States each year.1
Our objectives were to evaluate the predictive ability and accuracy of the AJCC 6th edition classification system, to compare the 5th and 6th edition classification systems, and to discern the relative contributions of T, N, and M classifications to the overall predictive ability of the pancreatic cancer staging system.
MATERIALS AND METHODS
By using the NCDB (1992–1998), patients were identified on the basis of the International Classification of Diseases for Oncology, 2nd edition (ICD-O-2) for tumors of the exocrine pancreas: C25.0, C25.1, C25.2, C25.3, C25.7, C25.8, and C25.9. Histology ICD-O-2 codes were used to select patients with pancreatic adenocarcinoma. The most current diagnosis year with complete 5-year follow-up data was 1998.
Patients who underwent pancreatectomy were identified based on the Commission on Cancer's Registry Operations and Data Standards (ROADS) site-specific procedure coding.3 Pancreatectomy includes pancreaticoduodenectomy (with or without pylorus preservation), distal pancreatectomy, total pancreatectomy, and pancreatectomy not otherwise specified. Patients who underwent pancreatectomy are contrasted to those who had “no resection.” Palliative procedures and exploratory surgery without a pancreatectomy cannot be differentiated in the NCDB, and these patients are included with those who did not undergo resection or pancreatectomy.
Patients from 1992–1998 were restaged based on AJCC 6th edition guidelines (Table 1). The NCDB requires reporting of clinical and pathologic stage, and a combination of the 2 was used for patients who did not undergo pancreatectomy. The NCDB is the only national tumor registry that reports data on AJCC stage. Classification of nodal (N) and metastatic (M) disease did not change over this time period. There were changes in the definition of T classification over time (Table 2). In the 3rd and 4th editions of the AJCC staging manuals, T1-classified disease was divided into T1A and T1B. These definitions generally correlate with 6th edition T1 and T2, respectively. In the 2nd, 3rd, and 4th editions, T2 classification correlates with 6th edition T3. In the 2nd, 3rd, and 4th editions, the most advanced T classification was T3, and it correlates with 6th edition T4.
|Definitions of TNM|
|TX||Primary tumor cannot be assessed|
|T0||No evidence of primary tumory|
|Tis||Carcinoma in situ|
|T1||Tumor limited to the pancreas, 2 cm or less in greatest diameter|
|T2||Tumor limited to the pancreas, greater than 2 cm in greatest diameter|
|T3*||Tumor extends beyond pancreas but no involvement of celiac axis or superior mestenteric artery|
|T4*||Tumor involves the celiac axis or the superior mestenteric artery (unresectable)|
|NX||Regional nodes cannot be assessed|
|N0||No regional lymph node metastasis|
|N1||Regional lymph node metastasis|
|MX||Distant metastasis cannot be assessed|
|M0||No distantant metastasis|
|Stage 0||Tis||N0||M0||Localized within pancreas|
|Stage IA*||T1||N0||M0||Localized within pancreas|
|Stage IB*||T2||N0||M0||Localized within pancreas|
|Stage IIA||T3||N0||M0||Locally invasive, resectable|
|Stage IIB*||T1,2, or 3||N1||M0||Locally invasive, resectable|
|Stage III*||T4||Any N||M0||Locally advanced, unresectable|
|Stage IV||Any T||Any N||M1||Distant metastases|
|6th Ed (2003)||5th Ed (1998–2002)||4th Ed (1993–1997)||3rd Ed (1989–1992)||2nd Ed (1988–1985)||1st Ed (<1985)|
|Localized within pancreas||T0||T0||T0||T0||T0||T0|
Observed 5-year survival rates according to AJCC 6th edition staging were calculated on the basis of the date of diagnosis (because both operative and nonoperative patients were being studied) to the date of death or last follow-up. Survival was estimated by using the Kaplan-Meier method and compared by using the log-rank test.4 A concordance index was calculated to evaluate the discriminatory power of the staging system by the Department of Epidemiology and Biostatistics at Memorial Sloan Kettering Cancer Center as described previously.5, 6 This estimate measures the level of agreement between the staging system and actual outcome. The concordance index provides the probability that patients with worse expected outcomes (advanced stage) die earlier than patients with better outcomes (earlier stage). A value of 0.5 indicates chance alone is as predictive as the staging system; whereas, a level of 1.0 signifies perfect concordance. A Cox proportional hazards model was used with stage as the only variable, and the concordance index was calculated as a function of the Cox model.
Cox proportional hazards modeling was used to asses the relative impacts of T, N, M classification on survival while adjusting for potential confounding variables including sex, age (<55 years, 55–65 years, 66–75 years, 76–85 years, >85 years), and race (White, Black, Asian, Hispanic, Other).7 The level of statistical significance was set to P < .05. Statistical analysis was performed by using SPSS, version 14.0 (SPSS, Chicago, Ill) and SAS version 9.1 (SAS Institute, Cary, NC).
Of the 127,779 patients with pancreatic adenocarcinoma in the NCDB from 1992–1998, analysis was limited to 121,713 patients who had complete staging and follow-up information. Of those patients, 21,512 patients underwent pancreatectomy, and 100,201 were treated nonoperatively. Patient characteristics are listed in Table 3.
|No. of patients||All patients 121,713||Pancreatectomy 18,743 (15.4%)||No resection 102,970 (84.6%)|
For all patients with pancreatic adenocarcinoma, there was good survival discrimination by stage (P < .0001; Table 4; Fig. 1A). Median survival was 4.4 months for all patients. The concordance index evaluating the agreement between the 6th edition staging guidelines and actual outcome was 0.631 (95% confidence interval [CI], 0.614–0.647) for all patients. In patients who did not undergo cancer directed pancreatectomy (n = 100,201), the AJCC 6th edition classification system predicted stage-specific survival (Table 4; Fig. 1B). For patients who did not undergo pancreatectomy, the concordance index was 0.596 (95% CI, 0.579–0.614).
|All patients||No. of patients (%)||Observed survival||Median survival (mo)|
|Stage IA||5298 (4.4%)||44.4%||24.6%||18.5%||15.3%||13.6%||10.0|
|Stage IB||6662 (5.4%)||40.6%||22.0%||16.1%||13.0%||11.7%||9.1|
|Stage IIA||12,332 (10.1%)||36.1%||16.2%||10.2%||7.7%||6.5%||8.1|
|Stage IIB||14,398 (11.8%)||40.9%||16.5%||9.6%||6.7%||5.1%||9.7|
|Stage III||15,831 (13.0%)||30.2%||9.5%||4.8%||3.5%||2.7%||7.7|
|Stage IV||67,192 (55.2%)||8.8%||2.5%||1.4%||0.9%||0.7%||2.5|
|Stage IA||3412 (4.4%)||29.2%||10.5%||6.2%||4.6%||3.8%||6.8|
|Stage IB||4298 (5.4%)||26.0%||9.4%||5.7%||4.0%||3.4%||6.1|
|Stage IIA||8486 (10.1%)||25.0%||7.7%||4.1%||2.8%||2.4%||6.2|
|Stage IIB||6570 (11.8%)||26.9%||7.7%||3.8%||2.6%||2.0%||6.7|
|Stage III||12,981 (13.0%)||27.0%||7.3%||3.4%||2.4%||1.8%||7.2|
|Stage IV||64,454 (55.2%)||8.3%||2.3%||1.2%||0.8%||0.6%||2.5|
|Stage IA||1886 (8.8%)||71.3%||50.2%||40.7%||34.7%||31.4%||24.1|
|Stage IB||2364 (11.0%)||67.3%||45.4%||35.3%||29.6%||27.2%||20.6|
|Stage IIA||3846 (17.9%)||60.7%||34.9%||23.8%||18.4%||15.7%||15.4|
|Stage IIB||7828 (36.4%)||52.7%||23.8%||14.4%||10.2%||7.7%||12.7|
|Stage III||2850 (13.2%)||44.5%||19.3%||11.0%||8.1%||6.8%||10.6|
|Stage IV||2738 (12.7%)||19.2%||8.4%||5.3%||3.7%||2.8%||4.5|
In patients who underwent pancreatectomy (n = 21,512), the classification system adequately predicted stage-specific survival (P < .0001; Table 4; Fig. 1C). The median 5-year survival by stage was IA 24.1 months, IB 20.6 months, IIA 15.4 months, IIB 12.7 months, III 10.6 months, IV 4.5 months (P < .0001). The concordance index for patients who underwent pancreatectomy was 0.613 (0.579–0.614).
In comparison to the 5th edition staging system, the 6th edition resulted in similar survival discrimination except for stage I patients who were divided into stages IA and IB (Fig. 2). The remaining changes from the 5th to the 6th edition were simply a reorganization of the existing stage groups: stage IVa in the 5th edition was changed to stage III in the 6th edition, and stage III in the 5th edition was changed to stage IIB in the 6th edition. The concordance index for the 5th edition was 0.630 (95% CI, 0.613–0.647) for all patients, 0.612 (95% CI, 0.565–0.660) for resected patients, and 0.596 (95% CI, 0.579–0.614) for nonsurgical patients. The concordance indices of the 5th edition were statistically similar to that of the 6th edition.
To determine the independent impact of T, N, and M classifications on survival, we examined these factors in a multivariate model while adjusting for sex, age, and, race (Table 5). In examining all patients with pancreatic cancer, tumor size and distant metastases had more impact on predicting survival than nodal status. However, in patients who underwent pancreatectomy, tumor size, nodal status, and distant metastases all affected the likelihood of death. Moreover, the likelihood of death increased as the T classification increased. In patients who did not undergo pancreatectomy, distant metastasis was the factor that had the most influence on survival.
|Hazard ratio (95% confidence interval)|
|All patients||Pancreatectomy||No resection|
|T1||1.00 (Referent)||1.00 (Referent)||1.00 (Referent)|
|T2||1.20 (1.16–1.24)||1.21 (1.13–1.29)||1.15 (1.11–1.20)|
|T3||1.23 (1.19–1.27)||1.42 (1.34–1.51)||1.14 (1.10–1.18)|
|T4||1.46 (1.41–1.50)||1.86 (1.73–1.99)||1.15 (1.11–1.20)|
|N0||1.00 (Referent)||1.00 (Referent)||1.00 (Referent)|
|N1||1.08 (1.06–1.10)||1.41 (1.36–1.46)||1.11 (1.09–1.13)|
|M0||1.00 (Referent)||1.00 (Referent)||1.00 (Referent)|
|M1||2.51 (2.47–2.55)||2.19 (2.05–2.34)||2.04 (2.00–2.08)|
The AJCC staging system for pancreatic carcinomas has never been validated with data from a large cancer registry. Our objectives were to evaluate the 6th edition AJCC staging system for pancreatic adenocarcinoma, to compare changes from the 5th to the 6th edition, and to ascertain the relative significance of T, N, and M classifications in predicting survival. The results of this study demonstrate that the AJCC 6th edition staging scheme for pancreatic cancer provides sufficient survival discrimination. The staging system for pancreatic cancer is an effective predictor of survival for patients staged clinically and pathologically. Moreover, these data support revisions to the AJCC pancreatic staging guidelines that emphasize the importance of surgical resectability.
Our first objective was to evaluate the predictive ability of the 6th edition staging system. Prior institutional studies have estimated the AJCC 6th edition concordance index to be 0.59 in patients who underwent resection for pancreatic adenocarcinoma.8 Our concordance index was statistically similar at 0.613 for patients undergoing pancreatectomy. The Memorial Sloan Kettering Cancer Center's (MSKCC) Postresection Pancreatic Carcinoma Nomogram for Disease-Specific Survival has been reported to have a concordance index of 0.62.8 Thus, our data indicate that in comparison with their nomogram, the AJCC 6th edition staging system has similar predictive ability. However, both the TNM classification and the MSKCC nomogram have relatively limited discriminatory ability, both with a concordance index of <0.8.9 Nevertheless, nomograms have shown promise and will likely become more valuable as biochemical and molecular markers are incorporated.
The single discrepancy that we observed in the staging system was relatively minor. In the first 2 years after diagnosis in the nonsurgical patients, stage IIB disease showed better survival than IIA. This may be because of inconsistent clinical TNM classification in patients who do not undergo complete, pathologic staging. The difference between IIA and IIB (nodal involvement) is difficult to discern on computed tomography and likely is irrelevant in predicting survival in patients who have been not been selected to undergo curative resection.
Our second objective was to compare the 5th and 6th edition staging systems. Recent evidence-based investigation into the staging system for gallbladder cancer has demonstrated that changes from the 5th to the 6th edition cancer staging manual have resulted in a less predictive staging system.5 There were 2 modifications to the pancreatic cancer staging system from the 5th to the 6th edition. First, T3 and T4 were redefined to be more relevant to surgical decisions. The most ominous finding that makes a pancreatic tumor unresectable concerns invasion of the celiac axis or superior mesenteric artery, and the new T-classification definition attempts to clarify the distinction between a potentially resectable (T3) and an unresectable tumor (T4). This was a change in definition, and patient survival was accurately predicted in our study by T classification. The 2nd change in the AJCC 6th edition pancreatic cancer staging system was the shift of T1-T3 N1 from stage III to stage IIB. This change was prompted by the observation that lymph node-positive patients identified after complete resection and lymph-node sampling had better outcomes than patients who did not undergo complete resection. We did not find a significant difference in the concordance indices between the 5th and 6th edition staging systems. This is likely because the changes from the 5th to the 6th edition were mostly a reorganization of the stage classifications. Only stage I was broken into IA and IB, and although there was good discrimination between these 3, stage I and II in the 5th edition also had reasonable survival discrimination. Thus, a significant improvement in the concordance index was not expected for the 6th edition.
Our final objective was to determine the relative contributions of T, N, and M classification to the predictive ability of the overall staging system. For patients who underwent pancreatectomy, tumor size, nodal status, and distant metastases affected survival. Moreover, if additional biochemical or molecular markers are added to the staging system in the future, demonstration of a similar effect in a multivariate model would give credence to the new factor's inclusion in the staging system. For patients who did not undergo pancreatectomy, distant metastases had the largest impact on the likelihood of death. However, tumor size and nodal status also had an effect, although less pronounced. T, N, and M classification affect survival for resected and unresected patients providing further evidence that the TNM classification system provides appropriate survival discrimination.
Although staging systems will need to be modified as advances are made in our understanding of disease, frequent alterations to AJCC staging classifications lead to confusion among healthcare workers as multiple versions of the staging system may be in use simultaneously. In addition, altering staging systems frequently may not allow adequate time for evidence-based verification of the system. For example, the AJCC 6th edition was published in 2002, so the earliest possible patients staged with 6th edition received their care in 2003. Five-year survival data will not be processed and available until late 2009 at the very earliest. Thus, there is no possible way for the AJCC to base its changes on survival data from patients who were originally staged with 6th edition guidelines. The best possible alternative is to apply the new staging system retrospectively to a large national cancer registry, as we have done here. A potential limitation of this approach is that patients must be restaged retrospectively. This may introduce some inconsistencies, as the staging system has changed over time and not all patients can be backwardly converted. This is particularly the case with patients staged by using the 2nd edition AJCC Cancer Staging Manual. Thus, we excluded patients diagnosed before 1992.
The availability of large databases provides an excellent, nationally generalizable opportunity to validate the current system and to test proposed alterations. Each cancer site in the cancer staging manual should contain validated prognostic information for clinicians. Stability of staging systems over time will allow for validation. Outcomes research will benefit from these tempered, well tested, and judicious changes to staging classifications. In conclusion, the AJCC 6th edition pancreatic cancer staging system provides appropriate survival discrimination.
- 1American Cancer Society, Cancer Facts and Figures. http://www.cancer.org/docroot/STT/content/STT_1x_Cancer_Facts-Figures_2007.asp. Accessed January 21, 2007.
- 2AJCC Cancer Staging Manual. 6th ed. Chicago, Ill: Springer; 2002.
- 3Standards of the Commission on Cancer Volume II: Registry Operations and Data Standards. Vol II. Chicago, IL: Commission on Cancer; 1998.
- 6Concordance Probability and Discriminatory Power in Proportional Hazards Regression. Memorial Sloan Kettering Cancer Center Working Paper Series. New York, NY: The Berkley Electronic Press; 2005., .
- 7Regression models and life tables. J R Stat Soc. 1972; 34: 187–220..
- 9Regression Modeling Strategies. New York: Springer; 2001..