Fax: (813) 745-7229
Histologic characteristics enhance predictive value of American Joint Committee on Cancer staging in resectable pancreas cancer†
Article first published online: 22 JUL 2009
Copyright © 2009 American Cancer Society
Volume 115, Issue 18, pages 4080–4089, 15 September 2009
How to Cite
Helm, J., Centeno, B. A., Coppola, D., Melis, M., Lloyd, M., Park, J. Y., Chen, D.-T. and Malafa, M. P. (2009), Histologic characteristics enhance predictive value of American Joint Committee on Cancer staging in resectable pancreas cancer. Cancer, 115: 4080–4089. doi: 10.1002/cncr.24503
Appreciation is expressed to Debi Mote, BBA, for assistance in development of the study database.
- Issue published online: 4 SEP 2009
- Article first published online: 22 JUL 2009
- Manuscript Accepted: 24 FEB 2009
- Manuscript Revised: 6 FEB 2009
- Manuscript Received: 10 DEC 2008
- Gastrointestinal Cancer Research Fund of the Moffitt Cancer Center
American Joint Committee on Cancer (AJCC) anatomic stage group is considered relatively nondiscriminatory for predicting differences in survival after pancreatectomy for ductal adenocarcinoma, a perception confirmed in the authors' patients and by other reports. The authors' aim was to investigate the potential for improving the predictive value of AJCC staging by incorporating individually predictive histologic features into AJCC tumor-node-metastasis classification of anatomic extent, and determine the simplest combination of tumor characteristics predicting survival.
The authors determined survival of 137 patients who underwent pancreatectomy for ductal adenocarcinoma with curative intent (stage Groups IA - IIB) at Moffitt Cancer Center during the last 2 decades using data obtained from medical record review, the Moffitt Cancer Registry, and the Social Security Death Index. Histologic characteristics were confirmed by expert review.
Median survival was 21.2 months after pancreatectomy with a 3-year disease-specific survival of 36%. Univariate Kaplan-Meier analysis and multivariate Cox proportional hazard modeling found worse survival with local extrapancreatic extension, poorly differentiated histology, and lymphatic invasion within tumor (P < .05). Survival was not worse with nodal metastases, microscopically positive resection margins, and perineural or venous invasion, nor was survival better with cancer arising from an intraductal papillary mucinous neoplasm. Kaplan-Meier estimates for different variable combinations showed prognosis was best for well- or moderately differentiated tumors without lymphatic invasion and confined to the pancreas (9.9 years median survival), worst for poorly differentiated tumors with lymphatic invasion and local extension beyond the pancreas (8.5 months median survival), and intermediate for well- or moderately differentiated tumors with either lymphatic invasion or local extension beyond the pancreas (21.2 months median survival).
A simple combination of tumor differentiation, lymphatic invasion within the tumor, and local extrapancreatic extension predicts survival after pancreatectomy for ductal adenocarcinoma. Cancer 2009. © 2009 American Cancer Society.
More than 37,000 new cases of pancreatic cancer are expected in the US in 2008, with the great majority of these being ductal adenocarcinomas.1 No greater than 5% of all pancreatic cancer patients will survive 5 years, the lowest long-term survival rate of any cancer. The only real chance of long-term survival with pancreatic adenocarcinoma belongs to those patients with localized disease in whom a potentially curative resection can be performed. Yet even among patients who undergo pancreatectomy with curative intent, the outcome is highly variable, with survival ranging from a matter of months to many years.2
The practice of classifying various kinds of cancers by stage arose from the observation that survival was better in patients with “early” localized cancers than for “late” cancers extending beyond the organ of origin. The American Joint Committee on Cancer (AJCC) tumor-node-metastasis (TNM) classification system is based on the premise that cancers with the same anatomic extent and histology will share similar behaviors and outcomes.3 In the AJCC system, tumors are classified by 3 significant events in the life history of a cancer: local tumor growth (T), spread to regional lymph nodes (N), and distant metastases (M). T, N, and M classifications are combined into stage groups intended to describe the anatomic extent of the cancer in a way related to its natural history. However, AJCC anatomic stage group (IA, IB, IIA, IIB) is considered relatively nondiscriminatory for predicting differences in survival among patients who have undergone pancreatectomy for resectable ductal adenocarcinoma,4, 5 a perception confirmed by analysis of outcomes in our patients and the work of others.6
For pancreatic adenocarcinoma, as for most other cancers, AJCC staging is concerned with only the anatomic extent of the disease.3 However, the AJCC does recognize that histologic characteristics of a tumor may be significant determinants of prognosis in some cancers, such as the histologic grade in soft-tissue sarcoma. Various histopathologic characteristics have been reported to be individually predictive of survival after pancreatectomy for ductal adenocarcinoma.4-21 Our aim in this study was to incorporate individually predictive histologic characteristics into the present TNM system of classification by anatomic extent and determine the simplest combination of these features that is predictive of survival after pancreatectomy for ductal adenocarcinoma.
MATERIALS AND METHODS
We previously reported our experience with 238 patients who underwent pancreatic resection for a primary cancer or premalignant neoplasm at the H. Lee Moffitt Cancer Center and Research Institute from January, 1987 through December, 2006.2 The current study is focused on the 137 patients with stage IA-IIB ductal pancreatic cancer who survived at least 30 days following pancreatectomy with curative intent. At least 61 patients received adjuvant chemotherapy, radiation therapy, or both, while no one underwent neoadjuvant therapy. Patients were identified by searching surgical and pathology administrative databases, as well as the Moffitt Cancer Registry. A pathologist with expertise in pancreatic cancer (BAC) confirmed the histologic diagnosis by review of stained sections cut from fixed surgical specimens. The World Health Organization International Classification of Diseases for Oncology, third edition (ICD-O-3),22 was used to classify pancreatic neoplasms based on the dominant histologic appearance. Data from the Moffitt Cancer Registry, the patient's medical record, and the Social Security Death Index were reviewed to determine the patient's clinicopathologic characteristics, tumor stage, survival of each patient following surgery, whether the patient was deceased as of most recent knowledge, and if there was evidence of disease at that time. Disease stage for all patients was defined by the present-day TNM classification of the AJCC Cancer Staging Manual, sixth edition.3 This study was approved by the Institutional Review Board of the University of South Florida, Tampa.
Disease-specific survival was estimated by the Kaplan-Meier method, using death with evidence of disease as the endpoint. Median survival, as well as 1-, 2-, 3-, and 5-year probabilities of survival, were determined. The log-rank test for equality of survival functions was used to compare Kaplan-Meier survival curves. Multivariate Cox proportional hazards modeling was used to determine the hazard of death with evidence of disease posed by a given variable, at any given time after surgery, adjusting for all other covariates. Global and covariate-specific tests based on both residuals and on graphical methods showed no evidence for violation of the proportional hazards assumption in all instances. Proportions were compared by the chi-squared test.
The Kaplan-Meier disease-specific survival estimate for all 137 patients who underwent resection of stage IA-IIB ductal pancreatic cancer and survived at least 30 days after surgery is shown in Figure 1. Median survival was 21.2 months after resection, with the longest survival known to be 15.7 years without evidence of disease. The probabilities of surviving 1 year, 2 years, 3 years, and 5 years after resection were 0.70 (95% CI, 0.61, 0.78), 0.42 (95% CI, 0.32, 0.51), 0.36 (95% CI, 0.26, 0.46), and 0.32 (95% CI, 0.22, 043), respectively.
Table 1 shows the AJCC stage group, and T, N, and R classifications for all 137 patients who underwent pancreatectomy. The Kaplan-Meier survival estimates corresponding to each stage group are shown in Figure 2. The 3 survival curves representing the 128 patients with stages IB, IIA, and IIB cancers did not differ significantly from each other (P = .26). Survival for stage Group IA, which included only 9 patients, was significantly greater than the other stage groups taken together (P < .05). Median survival for stage IA was 9.85 years after resection, as compared with median survivals of 1.94 years, 1.57 years, and 1.43 years for stage Groups IB, IIA, and IIB, respectively.
|IA tumor ≤2 cm, confined to pancreas, no positive nodes||9 (7)|
|IB tumor >2 cm, confined to pancreas, no positive nodes||21 (15)|
|IIA local extrapancreatic extension without celiac axis or SMA involvement, no positive nodes||44 (32)|
|IIB regional node metastasis||63 (46)|
|T1 tumor ≤2 cm, confined to pancreas||14 (10)|
|T2 tumor >2cm, confined to pancreas||28 (20)|
|T3 local extrapancreatic extension without celiac axis or superior mesenteric artery involvement||95 (70)|
|N0 no nodal metastases||72 (53)|
|N1 regional node metastases||65 (47)|
|R0 negative resection margin||100 (73)|
|R1 microscopically positive resection margin||37 (27)|
Univariate Kaplan-Meier survival curves for T, N, and R classifications are shown in Figure 3A, 3B, and 3C, while the corresponding numbers of patients staged are given in Table 1. Differences in tumor size (T2 vs T1) did not affect survival as long as the tumor was confined to the pancreas (P = .28). Local extrapancreatic extension of the cancer without celiac axis or superior mesenteric artery involvement (T3), however, was associated with significantly worse survival compared with cancers of any size confined to the pancreas (P < .05). Survival in patients with regional nodal metastases (N1) fell short of being significantly worse than in patients who did not have nodal metastases (P = .15). Finding microscopic evidence of cancer at any resection margin (R1) was not associated with a significantly worse prognosis (P = .51).
Univariate Kaplan-Meier survival estimates for different histologic characteristics are shown in Figures 3D and 4, while the corresponding numbers of patients described by these curves are given in Table 2. Origination of invasive ductal adenocarcinoma from an intraductal papillary mucinous neoplasm (IPMN) was not associated with significantly better survival (P = .54). Survival curves for patients with moderately differentiated and well-differentiated cancers did not differ significantly (P = .47). Poorly differentiated histology, however, was associated with significantly worse survival compared with more differentiated histology (P < .05). Lymphatic invasion within the tumor was associated with a worse survival (P < .05), although as we have already noted, the presence of regional lymph node metastases fell short of being associated with a significantly worse prognosis. About 69% of patients with lymphatic invasion within the tumor had regional lymph node metastases, but in the absence of lymphatic invasion, only 7.5% of patients had lymph node metastases (P < .05). Thus, lymphatic invasion within the tumor appears to precede regional lymph node metastasis. Venous or perineural invasion within the tumor did not affect survival significantly (P = .44 and 0.47, respectively).
We used a multivariate Cox proportional hazard model to identify which histopathological variables individually predicted a worse survival when adjusted for all other covariates (Table 3). For this analysis, we combined all patients with a tumor of any size confined to the pancreas into 1 group (T1 and T2) because the survival curves for T1 and T2 were not significantly different. For similar reasons, we combined patients with well- or moderately differentiated tumors into 1 group. The multivariate Cox proportional hazard model identified local extrapancreatic extension (T3), poorly differentiated histology, and lymphatic invasion within the tumor as being the only significant predictors of worse survival, adjusting for other covariates. These variables are the same ones identified by the univariate Kaplan-Meier estimates as having predictive value. The hazard ratios indicate that death with evidence of disease is about 1.5 to 3.0 times as likely at any given time after surgery if any 1 of these 3 predictors alone is present.
|Tumor arose from IPMN|
|Histopathologic Characteristic||Hazard Ratio (95% CI)||P|
|T1 or T2 tumor confined to pancreas|
|T3 Local extrapancreatic extension without celiac axis or SMA involvement||1.53 (1.02-2.30)||.04|
|N0 no nodal metastases|
|N1 regional node metastases||.75 (.29-2.00)||.57|
|R0 negative resection margin|
|R1 microscopically positive resection margin||1.08 (.47-2.48)||.86|
|Tumor arose from IPMN|
|Well or moderately differentiated|
|Poorly differentiated||1.88 (1.23-2.87)||.003|
We next determined Kaplan-Meier survival estimates for different combinations of the 3 histopathologic characteristics that were identified by both univariate survival analysis and multivariate Cox modeling as being individually prognostic for survival. Only 115 of the 137 patients had information available for all 3 prognostic characteristics, and the distribution of these patients among the 8 possible combinations of characteristics is shown in Table 4. The majority of patients with poorly differentiated tumors had local invasion beyond the pancreas, as well as histologic evidence of lymphatic invasion within the tumor (Group 8). Median survival in this group of 13 patients was 8.5 months, with 1- and 2-year survival probabilities of 34% and 23%, and none known to live more than 27 months. Individual survival curves were not determined for combination Groups 5, 6, and 7, because the number of patients in each of these groups would be too small for the curves to be meaningful. Instead, we pooled the outcomes of all patients with poorly differentiated tumors, including 4 such patients in whom all 3 prognostic characteristics were not available, and the result is a single survival curve representing the outcome in a total of 26 patients with poorly differentiated tumors (Fig. 5). The outcomes of patients in Groups 2 and 3 were likewise pooled and a single survival curve determined, because the individual survival curves for Groups 2 and 3 were found to overlie each other and did not differ significantly (P = .37).
|Group||Histologic Differentiation||Local Invasion Beyond Pancreas||Lymphatic Invasion||No.|
|1||Moderate or well||No||No||15|
|2||Moderate or well||Yes||No||19|
|3||Moderate or well||No||Yes||14|
|4||Moderate or well||Yes||Yes||45|
Figure 5 shows that the prognosis became increasingly worse as the histopathologic characteristics of the tumor became reflective of a more aggressive tumor biology. Survival was significantly better for well- or moderately differentiated tumors confined to the pancreas and without histologic lymphatic invasion (Group 1), as compared with patients in other groups (P < .05). Survival was the worst for poorly differentiated tumors as compared with either the least histologically aggressive tumors (Group 1) or the intermediate group of patients with well- or moderately differentiated tumors who had either local invasion beyond the pancreas or lymphatic invasion (pooled Groups 2 and 3; P < .05). Interestingly, the survival curve for patients with moderately or well-differentiated tumors, local invasion beyond the pancreas, and lymphatic invasion (Group 4) was not significantly different from either the survival curve for poorly differentiated tumors (P = .27) or the curve for pooled Groups 2 and 3 (P = .29). Inspection of the survival curves of Figure 5 suggests that initially the Group 4 patients will have a poor prognosis similar to that of patients who have a poorly differentiated tumor, but after 1 year, the prognosis improves and the survival curve comes to overlie the curve for pooled Groups 2 and 3. Median survival and probability of survival at 1 year, 2 years, 3 years, and 5 years are shown for different combinations of histopathologic characteristics in Table 5.
|Probability of Surviving|
|No.||Median Survival, y||1 Year||2 Years||3 Years||5 Years|
|Moderate or well differentiated|
|Confined to pancreas, any size|
|No lymphatic invasion|
|Groups 2 and 3 combined||33||1.95||.81||.46||.38||.19|
|Moderate or well differentiated|
|Invasion beyond pancreas or lymphatic invasion|
|Moderate or well differentiated|
|Invasion beyond pancreas|
|May or may not have invasion beyond pancreas or lymphatic invasion|
AJCC TNM classification is used inconsistently for staging ductal adenocarcinoma of the pancreas, perhaps, in large part, because the prognosis of the disease is generally poor, with no >20% of patients being candidates for potentially curative surgery.11, 23 Resectable cancers are classified into group stages IA through IIB; however, AJCC staging does not adequately reflect the heterogeneity in outcomes among patients who undergo pancreatectomy. We confirmed the perception that AJCC anatomic staging is relatively nondiscriminatory for predicting differences in survival among patients who undergo pancreatectomy for resectable ductal pancreatic cancer.4, 5 We found that AJCC anatomic stage groups did not predict differences in survival among the stage IB, IIA, and IIB patients who accounted for 93% of our patients who underwent resection of pancreatic cancer with curative intent. Predictably, better survival was found in only the small number of patients with early stage IA cancers, that is, those less than 2 cm in size and confined to the pancreas without regional lymph node metastases.
We sought to improve upon AJCC anatomic staging by incorporating potentially predictive histologic characteristics into the TNM classification, an approach consistent with the AJCC's recognition that histologic features may be important determinants of prognosis in some cancers.3 In doing so, our goal was to separate patients into groups predictive of survival after pancreatectomy. The first step in our analytic strategy was to use Kaplan-Meier survival estimates to identify individually predictive histopathologic characteristics. Then Cox proportional hazards modeling was used to confirm which of these characteristics were predictive of survival after adjusting for confounding. Finally, Kaplan-Meier survival curves were determined for the different possible combinations of individually predictive histopathologic characteristics to separate patients into groups predictive of survival.
To determine which histopathologic characteristics we should test for being individually predictive of survival after pancreatectomy, we considered several factors to be important. First, an experienced pathologist should easily and reproducibly determine a predictive histopathologic characteristic. Furthermore, we believed that there should be a biologic rationale by which the characteristic might be expected to influence survival. Finally, we considered characteristics previously studied and the outcomes of these studies as reported in the literature. Although various histopathologic characteristics have been reported to be individually predictive of survival after pancreatectomy, not all studies are in agreement.4-21 Why this should be true is not entirely clear; however, the reasons in some cases may be as simple as the observation that not all studies evaluated the same characteristics or that characteristics were defined differently between studies. We found, as did at least some of the other studies, that poor tumor differentiation, histologic evidence of lymphatic invasion within the tumor, and local extrapancreatic invasion were individually predictive of a worse survival.4-21
Determining Kaplan-Meier survival estimates for the different possible combinations of the 3 individually predictive histopathologic characteristics allowed us to separate patients into groups predictive of survival after pancreatectomy. We were able to identify 4 groups of patients with distinctly different survival curves based on the histopathologic characteristics of their tumors. The observation that prognosis worsened as histopathologic characteristics reflected a more aggressive tumor biology lends credibility to our method and supports the validity of our findings. Of interest is the group of patients whose prognosis was initially similar to that of patients who had poorly differentiated tumors with the worst survival, but beyond a year, the survival curve came to overlie that of the patient group with an intermediate prognosis. The unique shape of the curve suggests that this group could be comprised of 2 groups of patients with different prognoses, although we do not have a more complete explanation.
We are not the first to use a combination of histopathologic characteristics to predict survival after pancreatectomy for adenocarcinoma. Taking an analytical approach similar to our own, a study from Japan used univariate Kaplan-Meier survival estimates to identify characteristics that were individually predictive of survival after pancreatectomy from among 20 largely histopathologic variables.18 Then Cox proportional hazard modeling was used to confirm that characteristics identified as being significant in the univariate analysis were significant predictors of survival in a multivariate analysis. The 4 characteristics found to be predictive of worsened survival were tumor size greater than 3 cm, the presence of tumor necrosis, and high semiquantitative scores for intrapancreatic neural and lymphatic invasion. Although lymphatic invasion was found to have predictive significance, regional lymph node metastasis did not, an observation in agreement with our results. Histologic differentiation was predictive only of disease-free survival in the univariate analysis in this study, whereas we found that poor differentiation was the dominant predictor of poor survival. It is at this point that the investigators' approach to analysis differed from ours, as they attempted to establish a predictive histologic classification based on how many of the 4 predictive characteristics were present. This scoring system assumes that each histologic characteristic is of equal predictive value, which is not necessarily the case. Kaplan-Meier survival curves stratified by number of predictive characteristics present showed that as the number increased, survival worsened progressively. In another study by Brennan and colleagues, the goal was to determine the best estimate of survival for any given patient based on that individual patient's own clinicopathologic characteristics, rather than to classify patients into groups that could be expected to have similar prognoses.4
In summarizing what we know about predicting survival after pancreatectomy for ductal adenocarcinoma, we begin with our confirmation of the perception that AJCC staging based on anatomic extent alone is an insufficient predictor. A number of studies have shown that various histopathologic characteristics are individually predictive of survival, but few studies have attempted to use a combination of variables to predict survival. Our analysis incorporated individually predictive histologic characteristics into a logical extension of the AJCC anatomic staging system based on tumor biology. We found that a simple combination of readily available histopathologic characteristics reflecting aggressiveness of tumor biology (differentiation, local extrapancreatic extension, and lymphatic invasion within the tumor) predict survival after pancreatectomy. The value of staging patients in this way goes beyond being able to predict how long an individual patient will survive after pancreatectomy. Based on staging by histopathologic characteristics, select patients in whom long-term survival is expected may elect to avoid adjuvant therapy so as to be spared its adverse effects. Clinical trials assessing the benefits of new adjuvant therapy regimens need to consider patient histopathologic characteristics in study design and take into account any differences between study arms. As we come to better understand the biology of pancreatic cancer, future studies may show that molecular staging is of value in predicting survival after pancreatectomy for ductal adenocarcinoma.
Conflict of Interest Disclosures
Supported in part by the Analytic Microscopy Core Facility of the Moffitt Cancer Center. Funded in part by the Gastrointestinal Cancer Research Fund of the Moffitt Cancer Center.
- 2Outcomes following resection of pancreatic adenocarcinoma: 20-year experience at a single institution. Cancer Control. 2008; 15: 288-294., , , et al.
- 3GreeneFL, PageDL, FlemingID, et al, eds. AJCC Cancer Staging Manual. 6th ed. New York: Springer-Verlag; 2002: 157-164.
- 22FritzA, PercyC, JackA, et al, eds. International Classification of Diseases for Oncology, 3rd ed. Geneva, Switzerland: World Health Organization; 2000.