The use of adjuvant chemoradiation for pancreatic adenocarcinoma (PAC) is accepted in North America, but there is a paucity of data to support this practice. The relation between adjuvant therapy and survival was assessed in a population-based cohort of patients with PAC.
A review was conducted of all cases of resected PAC from 1996 to 2003 using data from the state cancer registry augmented with data from primary medical record review. Use of adjuvant therapy was ascertained from registry data. Survival was assessed using the Kaplan-Meier method, and a Cox proportional hazards model was developed for multivariate analysis.
A total of 298 patients from 27 hospitals met criteria for inclusion. There were 228 patients (76.5%) who were resected with curative intent, with a median overall survival of 12 months. The 6-month, 1-year, and 5-year survival rates were 80.2%, 58.4%, and 6.7%, respectively. Of the 228 patients resected, 122 (53.5%) received adjuvant treatment and had a median survival of 13.0 months versus 11.0 months for those with no adjuvant treatment (P = .16). After adjustment for surrogates of performance status, significant predictors of overall survival included no weight loss, T1/T2 pathologic stage, a microscopically complete resection (R0), and receipt of adjuvant therapy.
Pancreatic cancer remains one of the most formidable challenges in clinical oncology. For 2009, the American Cancer Society (ACS) estimates there will be 42,470 new cases of pancreatic adenocarcinoma (PAC) and 35,240 deaths attributable to the disease.1 PAC accounts for only 3% of cancer-related diagnoses, but it is responsible for an estimated 6% of all cancer-related mortality.1
Resection remains the only potentially curative therapy for patients with PAC. Unfortunately, most patients that undergo resection eventually develop disease recurrence and die of disease. The need for effective adjuvant therapy is clear. Since the 1985 publication of the Gastrointestinal Study Group Trial (GITSG),2 adjuvant chemoradiotherapy with a 5-fluorouracil-based regimen has been the de facto standard for patients with good performance status in the United States. Retrospective case series from single institutions have provided support for this approach, but these studies are limited by issues of patient selection.3, 4 The selection of good performance status patients for adjuvant therapy over their medically fragile counterparts confounds the observed survival advantage of adjuvant treatment in these retrospective studies. Studies using administrative data have hinted at benefit from chemoradiotherapy.5, 6 However, these studies are also limited by issues of patient selection, and they lack the clinical detail to allow for accurate adjustment for case mix.
The objectives of this study were to measure the use of adjuvant therapy, particularly the use of chemoradiotherapy, within a large population-based cohort of patients treated by a variety of different surgeons and oncologists within a spectrum of healthcare settings. In addition, this study also sought to describe the outcome of patients treated with adjuvant chemoradiotherapy and to assess the potential contribution of adjuvant therapy to postoperative survival.
MATERIALS AND METHODS
Using the Oregon State Cancer Registry (OSCaR), we identified all cases of PAC diagnosed between January 1, 1996, and December 31, 2003. This time period reflects the inception of the cancer registry through the most recent update to the registry at the time our study was undertaken. The data from this project are the result of a combination of information abstracted from deidentified patient medical records and data collected by the OSCaR, resulting in a retrospective cohort study design.
OSCaR requested data from all hospitals in Oregon with cases of PAC during the study period which had an operative intervention. The hospitals provided the operative note, pathology report, discharge summary, and the admission history and physical pertaining to the PAC operation for each case. These records were then returned to OSCaR, sorted by the registry staff, and deidentified. The registry then released the records to the research team, and we abstracted variables of interest for each case. We constructed a comprehensive database, which merged clinical data with registry data using the Statistical Package for Social Sciences (SPSS, Version 13.0). These data were then merged with survival and treatment data contained in OSCaR using an encrypted patient identifier. Data available included patient demographics (eg, age, gender), presenting symptoms (eg, weight loss, back pain), tumor data (size, lymph node status, margins), use of adjuvant therapy (eg, radiotherapy, chemotherapy, and chemoradiotherapy), vital status, and survival. To allow adjustment for performance status and surgical recovery, we identified whether patients were taking an oral diet at discharge and recorded the duration of their surgical hospitalization in days. The hospital volume of pancreatic cancer resections was determined by the number of resections performed per year in each hospital. The hospitals were then divided into quartiles based upon this volume and given a categorical designation. The study was reviewed and approved by the Oregon Health and Science University (OHSU) Cancer Institute review board and the OHSU Institutional Review Board.
For study eligibility, we required that patients were diagnosed between 1996 and 2003 and that they had a confirmed pathologic diagnosis of PAC. All study patients were considered potentially resectable by a surgeon after preoperative imaging evaluation, and either a laparoscopy or laparotomy was then performed.
Patients were excluded from the analysis if they were taken to the operating room for a procedure with palliative intent (eg, choledochojejunostomy) or if pancreatic cancer was not suspected at the time of operation (eg, trauma laparotomy). We also excluded atypical histologies, including intraductal papillary mucinous neoplasms, lymphoma, sarcoma, and cystadenocarcinoma. In addition, we excluded patients with confirmed duodenal adenocarcinomas and ampullary cancers.
We used information extracted from the OSCaR database pertaining to patient survival. Uncensored cases were defined as those cases reaching the endpoint of interest (ie, death) and censored cases as those cases who survived beyond the end of the follow-up period or who were lost to follow-up. Overall survival (OS) time was calculated from the date of definitive cancer resection to the date of last follow-up. Cumulative event rates were calculated using the method of Kaplan and Meier (K-M).7 We performed univariate analyses using the log-rank test to compare differences between categorical groups and the Mann-Whitney U test for ordinal data. The variables were assessed for violations of linearity and whether stratification was necessary. We developed Cox proportional hazards models8 using several clinicopathologic variables to determine the association of each with OS. The models were built by a purposeful selection method using the Wald statistic as described by Hosmer and Lemeshow.9 Relative risks were expressed as hazards ratios (HRs) with a 95% confidence interval (95% CI). The final model was evaluated for multicollinearity by examining correlation matrices of regression coefficients, for violation of the assumption of proportional hazards by examining plots of partial residuals versus survival time, and for goodness-of-fit using the method proposed by May and Hosmer.9, 10 Significance levels were set at P = .05. All tests were 2-sided. All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS) (version 13.0; SPSS Inc, Chicago, Ill).
In total, 447 records were identified in the state registry matching the inclusion criteria of the study. Twenty cases were out-of-state and/or in the Veterans Affairs system and therefore not accessible. Of the remaining 427, the state registry did not receive 18 records for initial deidentification. In summary, 409 cases (91.5% of the total identified and 95.6% of the records requested) were deidentified by the state registry staff and released to us for data abstraction. Two hundred ninety-eight (72.9%) of the cases received satisfied the study inclusion criteria with the majority of cases being excluded because the histopathology indicated a malignancy other than adenocarcinoma (eg, neuroendocrine malignancies). We received records from 27 hospitals throughout the state; however, the patients satisfying the eligibility criteria came from 21 hospitals. Of the 298 patients, 240 (80.5%) were resected; of those 240 patients, 12 patients were excluded because their pathology reports revealed metastatic disease from intraoperative biopsies. In summary, 228 of the 298 (76.5%) patients undergoing surgical exploration were classified as undergoing a resection with curative intent.
The mean age of the study population was 65.6 years (standard deviation [SD], 10.7 years; range, 26-90 years) (Table 1). Men comprised 50.4% of the patient population. The majority (57.0%) of patients underwent resection between 2000 and 2003. The volume of pancreatic resections per year was defined as 0 to 3.5 resections indicates low, 3.6 to 6 indicates medium, and >6 indicates high. The most common presenting clinical symptoms were jaundice (70.6%), epigastric pain (55.7%), and weight loss (52.6%). Greater than 95% of patients had computed tomographic scanning evidenced by their preoperative notes. Endoscopic ultrasound (EUS) was used in 36.0% of the population and of those undergoing EUS a fine-needle aspiration was performed in 27.6%. Almost one-half (46.5%) of the patients had a biliary stent placed endoscopically before their surgery. The mean length of hospitalization was 14.8 days (SD, 10.4 days); 14 patients (6.1%) died during their hospitalization or within 30 days of their hospital discharge.
Table 1. Demographic and Clinical Characteristics of Patients With Resected Pancreatic Adenocarcinoma
The majority of the cases (82.5%) were malignancies of the head of the pancreas, neck, or uncinate process (Table 2). Overall, 28.9% of patients were laparoscopically explored before their planned resection and of the patients ultimately resected, 19.2% had undergone laparoscopic exploration. Of the 298 patients who were candidates for resection based on preoperative imaging modalities, 228 (76.5%) were resected with curative intent. A classic Whipple was performed in 43.4%, a pylorus-preserving pancreaticoduodenectomy was performed in 36.0%, and a total pancreatectomy in 3.1% of the patients. Of patients undergoing a pancreaticoduodenectomy (n = 181) 20.4% underwent a duct-to-mucosa reconstruction, 75.1% underwent pancreatic stump invagination, and the type of pancreaticojejunal anastomosis was unknown in 3.9%. Vascular resection and reconstruction of the portal vein, superior mesenteric vein, or confluence was performed in 6.1% of patients. Overall, 22.8% of patients required a blood transfusion. A jejunostomy feeding tube was placed in 51.3% of patients.
Table 2. Operative Characteristics of Patients With Resected Pancreatic Adenocarcinoma
The retroperitoneal margin (also known as the uncinate margin) was identified in the pathology report in 46 of the 188 (24.5%) of the relevant cases. As shown in Table 3, 158 (69.3%) of the patients underwent a microscopically complete resection (R0). The mean number of lymph nodes sampled was 7.8 (SD, 5.6 lymph nodes; range, 0-27 lymph nodes). The number of lymph nodes resected was stratified as displayed in Table 3; 52.7% of the patients had ≥6 lymph nodes removed as part of their surgical specimen. Overall, 69.7% of patients were classified with T3 disease, and 49.6% of the patients were classified as having N1 disease according to the American Joint Committee on Cancer (AJCC), sixth edition classification.11 The majority of patients were stage IIB (48.2%).11 Eight (3.5%) cases had no staging information available from the pathology report or from the state registry.
Table 3. Pathologic and Treatment Characteristics of Patients with Resected Pancreatic Adenocarcinoma
No. of Patients (N = 228)
AJCC indicates American Joint Committee on Cancer.
Positive surgical margins indicates R1 or R2 resection.
According to the AJCC Cancer Staging Manual, 6th ed.
Of the 228 patients who were resected and did not have metastatic disease, 89 (46.5%) did not receive any form of adjuvant treatment. Of the 122 patients who underwent adjuvant therapy, 89 (73.0%) underwent combined chemoradiotherapy. Information regarding the number of cycles or agents used in the chemoradiotherapy regimen was not available for analysis. There was no association noted between the volume of the treating institution and the use of adjuvant therapy. We were unable to determine the extent of radiation fields, fractionation schedules, or total doses.
The median OS of the 228 patients was 12 months (95% CI, 11-13 months). The 6-month, 1-year, and 5-year survival rates were 80.2%, 58.4%, and 6.7%, respectively. There were 58 (25.4%) censored cases. The K-M analysis is shown in Figure 1. Univariate analysis was performed with the log-rank test and univariate Cox modeling. The results of the analysis of OS are displayed in Table 4. Risk factors associated with a decreased OS in a univariate analysis included T stage (P < .001), positive surgical margin status (P = .01), AJCC stage (P < .001), and length of hospitalization following operation (P = .001). The receipt of adjuvant therapy did not reach statistical significance (P = .16) in a univariate analysis. Patients who underwent an R0 resection had a 13-month median survival compared with a 10-month median survival in patients who underwent an R1 or R2 resection (mean, 16 and 12 months, respectively). The survival by margin resection status is displayed in Figure 2.
Table 4. Univariate Analysis of Risk Factors Associated With Overall Survival
Multivariate analysis using the Cox proportional hazards model was performed for OS. We examined the following variables: age, gender, weight loss, back pain, cancer location, T classification, number of lymph nodes examined, N classification, margin status, grade of adenocarcinoma, year of surgery, hospital volume per year of pancreatic cancer resections, the length of hospital stay after pancreatic cancer resection, and the receipt of adjuvant therapy. In a multivariate purposeful backward selection method using the Wald statistic, 5 predictors of OS (Table 5) were identified: reported preoperative weight loss, T classification, N classification, margin status, and receipt of adjuvant therapy. The surrogate variables for patient performance status (length of hospitalization and whether the patient was tolerating a diet at the time of discharge) were not found to be statistically significant or influential in the modeling and therefore were not included in the final model. Although not significant at the P < .05 level, the inclusion of lymph node status into the final model significantly improved the performance and discriminative ability of the model. The strongest independent predictors of overall mortality was T3/T4 stage compared with T2/T1 with an HR of 1.66 (95% CI, 1.12-2.47) followed by positive margin status with HR of 1.60 (95% CI, 1.13-2.25). Patients who did not report weight loss preoperatively had improved survival compared with patients with weight loss. No receipt of adjuvant treatment was an independent predictor of mortality (P = .025), with patients who did not receive adjuvant treatment dying at approximately 1.5 times the rate compared with those that did receive treatment throughout the study period. We examined the correlation of the regression coefficients, and none of the variables were found to be highly correlated. The final model performed well on all computed and graphic goodness-of-fit measures.
Table 5. Multivariate Forward Stepwise Cox Regression Analysis: Final Model for Risk Factors for Worse Overall Survival in Patients with Resected, Nonmetastatic Pancreatic Adenocarcinoma, 1996 Through 2003
The aim of our study was to examine postoperative survival as well as the use and outcome associated with adjuvant therapy in patients with resected PAC in a population-based sample. To achieve these ends and to address the limitations inherent in large administrative data sets, we augmented the tumor registry data with an intensive review of the primary medical records for all patients.
Over the past decade, reports from several large institutions and analyses of administrative data sets have demonstrated that resection of PAC is performed with declining operative mortality, morbidity, and increasing long-term survival.3, 12-15 Our population-based study of resected PAC yields several notable findings. Our observation of a 12-month median OS and 6.7% 5-year survival resected pancreatic adenocarcinoma as observed on a population basis over an 8-year period is similar to the range of 5% to 15% that is reported in the literature.16-18
To assess the impact of the 58 censored cases in our analysis, we removed them and re-ran the survival analysis with the remaining 170 patients. The results, including the OS and median survival, were exactly the same. We believe the lower 5-year survival demonstrated in the current study reflects the outcome of patients with PAC treated across a variety of practice patterns and institutions. As such, our reported number is likely to be a more accurate reflection of population-based outcome for surgically treated PAC than outcomes as reported by single-institution series or from clinical trials. In a recent Surveillance, Epidemiology, and End Results (SEER) study19 the authors reported a 16.3% 5-year survival for resected PAC, which is substantially lower than that reported by large, single-institution series and closer to our population-based estimate. In another SEER study by Govindarajan et al,20 the reported survival at 5 years for patients with resected PAC was 15.7% with a median survival of 16 months, which is similar to the 12-month median survival observed in the current study. We believe a population-based analysis of survival for patients with resected PAC more accurately reflects the outcome of patients treated across variety of practice patterns in contrast to the survival reported from large, single-institution cancer centers.
The clinicopathologic factors found to be most predictive of survival in our population-based analysis were preoperative weight loss, T classification, margin status, and receipt of adjuvant therapy. Preoperative weight loss is likely a surrogate for systemic disease and poor overall nutrition status; this would appear to naturally correlate with decreased OS.
Our final survival model demonstrated that the AJCC T classification was significant, whereas the patient's lymph node status was not. We included the patient's lymph node status into our final model because it greatly improved the fit and discriminative ability of the model. In addition, when we developed our survival models, we explored several methods of lymph node status that have been recently reported in the literature to be significant in resected PAC, including the ratio of positive lymph nodes to total lymph nodes sampled, stratification of lymph nodes, and the total lymph nodes sampled.21 We did find any of the lymph node status renderings to be significantly correlated with survival in our analysis.
The impact of the surgical margin status on outcome has long been debated.22-24 The positive margin may be the result of technical features, disease biology, or patient selection. Irrespective of the causative mechanism, this study reaffirms the microscopically involved margin as a marker of poor outcome. Perhaps more importantly, our study demonstrates that there are often deficiencies in the pathologic assessment of the surgical margins. We found that the retroperitoneal margin, which is often involved, was identified in the pathology reports in only 46 of 188 (24.5%) of the relevant cases. Although the College of American Pathologists has developed a synoptic reporting format25 that will mandate examination of the retroperitoneal margin, this will remain an important quality improvement effort in pancreatic cancer care.
In our population-based study, the receipt of adjuvant therapy was associated with an increase in survival. Of the 228 patients with resected PAC, 39.0% underwent adjuvant chemoradiotherapy, 12.3% received adjuvant chemotherapy only, and 2.2% of patients underwent radiation only. These data do not indicate the doses or schedules of chemotherapy, nor the technical details of the radiation treatment. Despite general acceptance of the concept of adjuvant therapy for PAC during this study period, we found that 46.5% of the patients in our study that underwent curative resection did not receive any form of adjuvant treatment. This large percentage could be reflective of several factors. We reviewed our data to determine whether there was a detectable trend in the receipt of adjuvant treatment over time, given the recent improvements in adjuvant therapy for pancreatic cancer. There was not any significant trend over time or a suggestion of increased use of adjuvant therapy. In the final year of our study, 46.3% of patients did not receive any adjuvant treatment. It is possible that these data reflect the inability of state cancer registry to accurately capture receipt of adjuvant treatment. However, these data are consistent with recent literature,26 suggesting that up to 30% of patients do not receive adjuvant therapy after pancreaticoduodenectomy due to delayed recovery. Some studies have suggested a link between institutional volume of pancreatic cancer patients and use of adjuvant treatment. Aloia et al26 reported that patients with a good performance status who underwent resection at a high-volume institution, delayed recovery precluded receipt of adjuvant therapy in only 6% of patients. However, the data from the current study revealed that the volume of the treating institution was not associated with use patterns of adjuvant treatment.
The role of chemoradiotherapy after resection of pancreatic cancer remains controversial. The original GITSG trial, which demonstrated a survival advantage for adjuvant chemoradiotherapy, was a small study with a sample of 43 patients.2, 27 More recently, the European Study Group for Pancreatic Cancer (ESPAC-1) study demonstrated a survival advantage for chemotherapy but not for combined chemoradiation.18, 28 Although the trial has been criticized, it has prompted many European centers to shun radiation in favor of chemotherapy alone. This trend has gained momentum with the publication of a German study that indicated a survival advantage for adjuvant gemcitabine chemotherapy versus surgical resection alone.29 Despite the trends in Europe, concurrent chemoradiotherapy has remained the de facto standard for adjuvant treatment in the United States.30 The findings of the current study provide additional support for the use of adjuvant chemoradiotherapy in patients who are medically suitable for treatment.
Although this study draws on the diversity of population-based data supplemented by detailed record review, the study does have limitations. Due to the retrospective design of the study, there may be unmeasured clinical factors we could not capture that impacted the decision making of surgeons. We attempted to enhance the depth of our data set by augmenting abstraction of clinical data with data from the state cancer registry. With regard to the association between receipt of adjuvant therapy and increased survival, there undoubtedly is a bias toward healthier patients with better performance status receiving adjuvant therapy. We attempted to adjust for this by including the patient's ability to tolerate an oral diet at discharge and the length of hospitalization associated with the resection. These are imperfect surrogates for more rigorous measures of biologic and nutritional status such as serum albumin; however, they do account for the patient's status after pancreatic resection and incorporate the impact of hospitalization.
Although several studies31-33 have demonstrated an association between hospital case volume and postoperative adverse outcomes in pancreatic cancer surgery, we found no linkage in our study between hospital volume and survival. The absence of a volume effect may indicate that the overall quality of pancreatic surgery has improved across all practice venues; however, we must interpret these findings with caution. The geographic limitations of this study as well as a small sample when compared with administrative data sets are likely to preclude an accurate assessment of the volume to outcome relationship in postoperative survival for PAC.
Because our data on receipt of adjuvant treatment come from registry data, there is a possibility of underreporting in the state registry, which could influence our results. We do not have any external measure of validity because our chart review was restricted to the perioperative setting. However, the percentage of patients receiving adjuvant treatment in our study is similar to that reported in the analysis of large, national data sets.6, 34 Our finding of 53.5% of patients in Oregon receiving some form of adjuvant treatment is similar to the range observed by Bilimoria et al34 in their review of more than 300,00 patients in the National Cancer Database, suggesting we have adequate data capture and reporting. We found that 39% of patients received combination chemoradiotherapy, which is again similar to the 1995 to 2003 time period by Bilimoria et al34 demonstrating 38.7% of patients underwent combination therapy. In addition, their analysis demonstrates a survival advantage with adjuvant treatment, although their analysis lacks the clinical detail captured in our study.34 In a quality audit of the California tumor registry data for colorectal cancer, Ayanian et al35 found that the registry was 87% sensitive for the reporting of chemotherapy and 93% sensitive for radiotherapy. Although missing data regarding adjuvant treatment are always a concern in population-based analyses, such work by Ayanian et al35 indicates that, particularly for the receipt of radiation treatment, registry data are quite complete and accurate.
In conclusion, the results of the current study demonstrate a high proportion of patients resected after laparotomy on a population-basis. In our study, an R0 resection and receipt of adjuvant therapy were associated with increased survival. We believe that, although the data presented in this study are retrospective, they do support the practice that all patients with resected PAC should receive some adjuvant therapeutic program if clinically feasible within an appropriate time frame following resection. The recent CONKO-129 data affirm chemotherapy alone as a beneficial adjuvant program. The finding that less than half of the patients in this population-based study received adjuvant therapy indicates that broadening the use of adjuvant therapy in patients with resected PAC should be a target for quality improvement in the care of patients with this difficult disease. The contribution of combined modality chemoradiotherapy versus adjuvant chemotherapy alone remains an urgent clinical question in PAC. This is a question that must be addressed with well-designed, prospective clinical trials. However, in the near-term, these data support the careful evaluation of all patients with resected PAC for clinically appropriate adjuvant therapy.
We thank Andrew Glass, MD, the staff of the Oregon State Cancer registry, and Debra Erickson for their work to help complete this project.
CONFLICT OF INTEREST DISCLOSURES
Supported by a grant from the Medical Research Foundation of Oregon and the Tartar Trust Foundation.