Although chemoradiation often is administered as an adjuvant to pancreatic cancer surgery, recent reports have disputed the benefit of radiation therapy. The objective of this study was to determine the effect of adjuvant radiation therapy in patients with locally confined, lymph node-negative (N0) pancreatic cancer.
The Surveillance, Epidemiology, and End Results registry was used to identify patients who had undergone cancer-directed surgery for N0 pancreatic adenocarcinoma between 1988 and 2003. Kaplan-Meier survival curves were constructed to compare overall survival between patients who did and did not receive adjuvant external-beam radiation therapy (EBRT). Multivariate Cox regression analysis was used to determine the prognostic significance of EBRT when additional clinicopathologic factors were assessed. The analysis also examined the potential treatment selection bias of patients with survival <3 months.
A cohort of 1930 surgical patients with N0 disease was identified. The median survival was 17 months. Irradiated patients had significantly better survival compared with nonirradiated patients (20 months vs 15 months, respectively; P < .001). On multivariate analysis, adjuvant EBRT (hazard ratio [HR], 0.72; 95% confidence interval [95% CI], 0.63–0.82; P < .001), age, grade, tumor classification, and tumor location were independent predictors of survival. When patients with survival <3 months were excluded from the analysis, no difference in survival between the EBRT group and the nonradiation group was noted on univariate comparison (P value not significant). However, on multivariate analysis, EBRT remained an independent predictor of improved overall survival (HR, 0.87; 95% CI, 0.75–1.00; P = .044).
Despite improvements in surgical technique and the introduction of novel multimodal treatment regimens, the prognosis for patients with pancreatic cancer remains extremely poor. Currently, 5-year survival rates after curative resection for all patients range between 14% and 27%.1–9 Similar outcomes have been reported for lymph node-negative (N0) patients with a 5-year survival rate of 21% and a median survival of 20 months.10 Such dismal outcomes are caused in large part by high rates of locoregional and distant failure, even in patients with early-stage tumors.2, 11, 12 Given these patterns of recurrence, therapies that address both locoregional and systemic disease have been applied. Accordingly, adjuvant treatment regimens that incorporate chemotherapy, radiation, or a combination thereof have been implemented to improve survival further, beyond the expectations of surgery alone.13–16
Improved outcome with adjuvant chemoradiotherapy was demonstrated first in a randomized trial by the Gastrointestinal Tumor Study Group (GITSG).14 Recent trials conducted by the European Organization for Research and Treatment of Cancer (EORTC) and the European Study Group for Pancreatic Cancer (ESPAC-1) have shed doubt on the benefits of adjuvant chemoradiation.17, 18 Indeed, in the ESPAC-1 trial, poorer survival was noted with the administration of chemoradiation.18 In addition, the Pancreatic Cancer Meta-analysis Group recently reviewed patient data from several randomized trials.19 Their results did not identify a survival advantage or disadvantage for adjuvant chemoradiation; however, they did report an improvement in survival with adjuvant chemotherapy alone.19
Although there is supportive evidence for the use of adjuvant chemotherapy,18–21 the role of adjuvant radiation remains unresolved, and patients in the United States continue to receive combined therapy. The objective of the current study was to determine the benefit of adjuvant external-beam radiation therapy (EBRT) using a large U.S. population-based database with a study population that, to our knowledge, is far greater than the population in any currently available randomized trial or meta-analysis. We hypothesized that positive lymph node status may be a reflection of micrometastatic, systemic disease and that this subgroup is unlikely to benefit from additional locoregional therapy. We excluded these patients from our analysis and examined the relationship between adjuvant EBRT and survival in patients who had lymph node negative (N0) pancreatic cancer.
MATERIALS AND METHODS
The Surveillance, Epidemiology, and End Results (SEER) registry22 was used to identify patients who had undergone cancer-directed surgery for histologically confirmed, N0, invasive pancreatic cancer between the years 1988 and 2003. Specifically, SEER surgical codes that indicated excision of tumor or extensive pancreatic and multiorgan resections were included. Patients who underwent exploratory surgeries, biopsies, or lymph node dissections alone were excluded. Patients who had no lymph nodes or had an unknown number of lymph nodes removed also were excluded. Our initial query identified 2250 patients who satisfied these inclusion criteria. On review of the available histologies for pancreatic cancer, the overwhelming majority of tumors were coded as either adenocarcinoma or ductal carcinoma. The several remaining histologies (n = 275 tumors) comprised a group of patients who had a disproportionately longer survival and, thus, were excluded (data not shown).
Because T4 lesions (ie, tumors involving the celiac axis or superior mesenteric artery23) represent a heterogeneous group of tumors for which surgical indications are unclear, T4 lesions (n = 45 tumors) were also excluded from the current analysis. Because the SEER database did not define tumor (T) classification until 2004, the T classification was derived from the variable for “local extent of disease” by using criteria from the American Joint Committee on Cancer, 6th edition.23 The specific SEER local extension codes that were used to define the T4 group included 62 through 64, 67, 70, and 76 through 80. These codes correspond to the T4 classification in the updated 2004 database and represent tumors with known major vascular invasion and further contiguous extension. The additional exclusions led to a final count of 1930 patients who were included in our statistical analysis.
The primary outcome measure of interest was overall survival. The main prognostic factor of interest was the administration of adjuvant EBRT, which was defined as the administration of “beam radiation” after surgery, as coded by the SEER variables “radiation” and “radiation sequence with surgery.” Patients who received neoadjuvant radiation (alone or in combination with adjuvant EBRT), intraoperative radiation, adjuvant radiation in the form of radioisotopes or radioactive implants, and patients in whom the radiation sequence was unknown were included in the descriptive portion of the analysis but were excluded from comparative survival analysis. Other clinicopathologic information gathered from the database included demographic data (eg, age, sex), local extent of disease, grade, and tumor location.
Clinicopathologic characteristics of the adjuvant EBRT and nonradiation groups were compared by using independent t tests for continuous variables and chi-square tests for categorical variables. Overall survival estimates for the sample population and treatment subgroups were determined by using the Kaplan-Meier method. Differences in survival between tumors treated with or without adjuvant EBRT were compared by using the log-rank test. Then, multivariate Cox regression analysis was used to determine the prognostic significance of adjuvant EBRT, controlling for other potentially confounding clinical factors. The EBRT variable was coded as a dichotomous, categorical variable. Age was coded as a continuous variable, and indicator variables were used for T classification, grade, and tumor location in the final multivariate model. Effect modification by T classification, grade, and tumor location was determined using the likelihood-ratio test after adding the appropriate interaction terms to the nested multivariate model.
A concern that arises with this type of study is the potential for selection bias that is introduced by the inclusion of patients with very limited survival. These patients likely represent a subgroup with an increased incidence of short-term, perioperative morbidity and mortality in whom adjuvant therapy probably was not considered. Consequently, inclusion of these patients may lead to an overrepresentation of early operative deaths in the nontreated arm. To account for this bias, Kaplan-Meier and Cox regression analyses were repeated after excluding patients with survival <3 months (n = 164 patients). Of those 164 excluded patients, only 1 patient (0.6%) received adjuvant EBRT. The results are reported as hazard ratios (HR) and with 95% confidence intervals (95% CI) and as P values from the appropriate statistical tests of significance. P values <.05 were considered significant. The statistical analysis was performed using SPSS (version 12.0; SPSS Inc., Chicago, Ill) and STATA (version 8.0; Stata Corporation, College Station, Tex) software.
Table 1 presents the characteristics of the entire sample population. The mean age (±standard deviation) was 65.7 ± 10.7 years, and sex was distributed equally. The most common tumor location was the head of the pancreas, followed by the pancreatic tail and body; other locations comprised smaller fractions. Lesions tended to be locally extensive with the majority (65.0%) categorized as T3. Most tumors were either moderately or poorly differentiated; there were few well-differentiated or anaplastic lesions.
Table 1. Characteristics of the Sample Population (N = 1930)
SD indicates standard deviation.
Includes location coded as “pancreatic duct” by the Surveillance, Epidemiology, and End Results database.
Includes 13 patients who were treated postoperatively with radioisotopes, radioactive implants, or unknown radiation sources; excludes patients who received postoperative radiation in combination with neoadjuvant or intraoperative radiation.
Includes the patients who subsequently received additional postoperative radiation.
Approximately 40% of patients received radiation at some point during their treatment, with the overwhelming majority receiving adjuvant radiation therapy alone (Table 1). Thirteen of 674 patients in the adjuvant radiation group were treated with radioisotopes, radioactive implants, or an unknown radiation source. These patients were excluded from further analysis; thus, the final adjuvant EBRT group consisted of 661 patients.
Table 2 presents a comparison of clinicopathologic factors stratified by radiation group. With respect to tumor location, lesions of the pancreatic head and neck and of the pancreatic body and tail were combined because of the similar behavior and treatment of tumors in these sites and the relatively small sample sizes within the individual subgroups. In general, the characteristics of patients in the adjuvant EBRT and nonradiation arms were very similar with the exception of age; irradiated patients were younger than the nonirradiated group (age 67 years vs 63 years; P < .001) (Table 2).
Table 2. Comparison of Patient Characteristics by Radiation Group
The median survival for the entire study population was 17 months. Kaplan-Meier survival curves were constructed to assess survival after patients were stratified according to radiation treatment. On statistical comparison, the adjuvant EBRT group had significantly better survival compared to the nonradiation group (median 20 months vs 15 months, respectively; P < .001; Fig. 1).
On univariate Cox regression analysis, adjuvant EBRT (HR, 0.75; 95% CI, 0.67–0.84; P < .001), age at diagnosis, tumor location, grade, and T classification all were associated significantly with survival (Table 3). Multivariate regression analysis was performed with all prognostic factors included in the multivariate model. The results of the multivariate analysis are presented in Table 4. EBRT was associated with an approximately 30% decrease in the risk of death (HR, 0.72; 95% CI, 0.63–0.82; P < .001). Age, T classification, grade, and tumor location also were associated significantly with survival (Table 4). There was an approximately 1% increase in the risk of death per year of increasing age (HR, 1.01; 95% CI, 1.004–1.016; P < .001). The relationship between survival and T classification was linear in nature, and T3 tumors conferred the greatest risk of death (HR, 1.79; 95% CI, 1.44–2.23; P < .001). With respect to grade, patients who had moderately and poorly differentiated tumors had significantly worse outcomes compared with patients who had well-differentiated tumors. Although the tumor location variable also was associated with survival, the significance of this relationship resulted mostly from the disadvantage noted with the “overlapping” location, as coded by SEER. This finding reflects the generally worse prognosis observed in patients with more extensive local disease.
Effect modification of the radiation-survival association by the categorical variables (sex, T classification, grade, and tumor location) was examined by adding the appropriate interaction terms to the nested multivariate model. There was no significant effect modification by T classification, grade, tumor location, or sex, indicating that the effect of EBRT on survival did not differ significantly across the individual subsets defined by these variables (Table 4).
Survival Analysis After Exclusion of Early Deaths
Kaplan-Meier survival curves comparing radiation groups after the exclusion of early deaths are shown in Figure 2 and indicate that there was no significant difference in overall survival between patients who did and did not receive EBRT on univariate comparison (median 20 months vs 19 months, respectively; P = .14). The survival curves were stratified by T classification and indicated that there was no difference in survival with or without EBRT for patients with T1 lesions (median 26 months vs 32 months, respectively; P = .40) or T2 lesions (24 months vs 20 months, respectively; P = .16). However, overall survival for patients with T3 lesions was increased significantly with adjuvant EBRT (median 20 months vs 16 months; P = .005) (Fig. 3).
The results of multivariate analysis after the exclusion of early deaths appear in Table 5. Adjuvant EBRT was associated with improved survival after adjusting for the other covariates (HR, 0.87; 95% CI, 0.75–1.00; P = .044). Age, T classification, grade, and tumor location also were associated independently with survival and had correlations similar to those observed in the previous analysis (see Table 4). On interaction analysis, the survival benefit associated with adjuvant EBRT did not vary across different grades, sexes, tumor locations, and T classifications (all P for interaction >.05); however, univariate stratified survival analysis demonstrated a trend toward a better response for patients with T3 lesions (Fig. 3).
Table 5. Results of Multivariate Cox Regression Analysis With Early Deaths Excluded (n = 1634)*
Although recent reports have suggested a benefit with adjuvant chemotherapy, the role of adjuvant chemoradiation in patients with pancreatic adenocarcinoma remains unclear. The results of randomized trials have been conflicting.13, 14, 17–21 The first prospective study of adjuvant chemoradiation in pancreatic cancer was the GITSG trial, in which 22 patients were randomized into a surgery-alone arm and 21 patients were randomized into a treatment arm consisting of 2 20-Gy courses of EBRT (split-course schedule) combined with bolus and maintenance 5-fluorouracil (5-FU).13 A significant improvement in survival was demonstrated with adjuvant treatment compared with observation (20 months vs 11 months, respectively; P = .03). The results of this study, however, were met with criticism because of the small sample size and the potential introduction of treatment bias based on performance status.24
An EORTC trial was designed to address some of the drawbacks of the GITSG trial and eventually demonstrated a significant survival advantage with a similar regimen of combined 5-FU plus EBRT.17 Because the authors of that report included patients with primary biliary and duodenal cancers in their initial accrual, the subset analysis of pancreatic cancers alone was underpowered, resulting in a nonsignificant, 2-sided P value of .09 despite a clear divergence of the treatment group and control group survival curves.17 The appropriateness of the 2-sided trial hypothesis subsequently has been questioned in favor of a 1-sided design, which would have led to a significant 1-sided P value <.05.25
To our knowledge, the most recent and largest prospective randomized trial that addressed adjuvant treatment in pancreatic cancer was conducted by the ESPAC group. The ESPAC-1 trial used a 2 × 2 factorial design, which allowed comparisons of adjuvant chemoradiation, adjuvant maintenance chemotherapy, a combination of both, and observation alone.18 The administration of maintenance chemotherapy resulted in a significant improvement in survival compared with observation. However, the administration of chemoradiation was associated with a significant survival disadvantage. Although the study was not powered sufficiently to compare the 4 treatment subgroups individually, the same survival disadvantage with chemoradiation was apparent with or without maintenance chemotherapy.18
The disparate outcomes for adjuvant treatment regimens in the trials described above are difficult to explain. Unlike the GITSG study, the ESPAC-1 trial included patients with positive resection margins, a subgroup that may be more likely to benefit from the addition of locoregional treatment. However, locoregional treatment was detrimental in the ESPAC-1 trial. Moreover, the decreased survival noted in the chemoradiation arm was not caused entirely by treatment toxicity, as would be expected. According to the authors, all but 12 of the 237 deaths in the ESPAC-1 trial were cancer-specific and were unrelated to treatment-associated complications. It has been suggested that the observed decrease in survival with chemoradiation may have been caused by a delay in the administration of maintenance chemotherapy.18 However, this explanation does not account for the same pattern of decreased survival observed with chemoradiation in the subset of patients who were not administered maintenance chemotherapy.
The recent review published by the Pancreatic Cancer Meta-analysis Group attempted to clarify the role of adjuvant chemoradiation and chemotherapy in the setting of resectable pancreatic cancer.19 That study examined individual patient data from several randomized trials, including the GITSG, EORTC, and ESPAC-1 trials. The review also included previously unpublished data on 261 patients from the ESPAC group that were not part of the initial trial's 2 × 2 factorial design. That meta-analysis demonstrated a significant benefit for chemotherapy without any discernible advantages for chemoradiation.19 However, the meta-analysis was biased substantially toward the ESPAC-1 results given that trial's disproportionately large patient cohort. In a comparison of the individual studies, the ESPAC-1 data were contradictory to the findings of the GITSG and EORTC trials. With respect to the effect of chemoradiation, the authors of the meta-analysis also noted heterogeneity between the ESPAC-1 results and the previously unpublished ESPAC data.19
The results of the ESPAC-1 trial are not accepted widely in the United States because of the difficulties in interpreting the results and because the treatment protocols differ from the regimens that are used currently.16, 26 The ESPAC-1 trial used split-course EBRT to a total dose of 40 Gy with concurrent bolus injections of 5-FU, whereas the EORTC trial used the same split-course EBRT schedule with concurrent, infusional 5-FU. Most centers in the United States employ a continuous course of EBRT with total doses from 45 Gy to 54 Gy in fractions from 1.8 Gy to 2 Gy.16, 26 Although this regimen has not been investigated in a randomized study for pancreatic cancer, it has been shown to be more effective radiobiologically in other malignancies than a moderate-dose, split-course regimen.27 Additional methodological concerns regarding the delivery of radiation therapy remain. Although the ESPAC-1 protocol provided guidelines for total dose and fractionation, treatment volume was not specified. Therefore, it is unclear whether radiation treatment, toxicities, and subsequent outcomes were consistent across institutions.
Given the equivocal reported outcomes with adjuvant radiation, our objective was to assess its role in the treatment of pancreatic cancer using a large U.S. population-based registry. We limited the study to patients with early-stage N0 tumors. These patients are less likely to have undetected metastatic disease and may sustain more benefit from adjuvant EBRT. The overall survival of 17 months in this population was consistent with published reports.10, 28 Our results demonstrated a survival advantage in patients who received adjuvant EBRT compared with nonirradiated patients. This advantage persisted after adjusting for several relevant clinical factors, such as age, T classification, grade, and tumor location.
A unique problem that arises with retrospective studies that examine survival with adjuvant interventions is the selection bias introduced by the inclusion of early deaths related to surgical morbidity or rapid progression of disease. This concern was addressed at least in part in the second portion of our statistical analysis with the exclusion of patients with survival <3 months. The radiation-survival correlation, as expected, was not as robust after this adjustment. However, on multivariate analysis, adjuvant EBRT still was associated independently with improved survival after accounting for other relevant prognostic factors. The discrepancy between univariate and multivariate findings may have been related to a slightly greater incidence of higher T classification lesions in the irradiated group (Table 2). The multivariate analysis was adjusted for this disparity.
An additional observation that can be made from the stratified survival data presented by the Pancreatic Cancer Meta-analysis Group is the apparent trend toward improved survival with chemoradiation in patients with margin-positive and poorly differentiated tumors, although the relationships did not reach statistical significance.19 The issue of effect modification by tumor grade, T classification, and location also was addressed in our current study. There was a significant univariate response with T3 lesions when early deaths were excluded (Fig. 3). These results suggest that, when comparing patients who have tumors with equivalent local extension, grade, age, and location, adjuvant EBRT confers a survival benefit. In addition, the survival benefit associated with adjuvant EBRT appears to be consistent across different grades, sexes, tumor locations, and T classifications on multivariate analysis.
Although every effort was made to account for statistical bias and to control for confounding factors, our current results should be interpreted in light of inherent problems with all registry studies as well as limitations specific to SEER. The SEER registry does not collect data regarding chemotherapy.21 However, past and current treatment practices in the United States suggest that most patients who are treated with adjuvant EBRT also receive chemotherapy.10, 13, 26, 27, 29–31 It is plausible that the survival advantage in our study may be the result of treatment with adjuvant chemotherapy rather than EBRT. In addition, the SEER registry does not collect information regarding resection margin status. Whether the survival difference we have demonstrated was associated with margin status cannot be determined. Finally, SEER does not provide EBRT details, such as total dose, fractionation schedule, and field size, and these factors have potential to influence outcome. Although the radiation benefit we have demonstrated was small and may have been caused by unidentified bias in the SEER database, the lack of a survival disadvantage with adjuvant EBRT also is compelling and is less likely the result of statistical or methodological biases. Given the benefit demonstrated with chemotherapy, it is unlikely that a trial with a radiation-only arm ever will be implemented, and a registry study of this nature may provide the best insight into the potential benefit of adjuvant EBRT.
In conclusion, there appears to be a small but significant benefit in survival with the administration of adjuvant EBRT for patients who have N0 pancreatic cancer in the U.S. population. This survival advantage is independent of several established prognostic factors for outcome. These finding are concordant with results from the GITSG and EORTC trials but are contradictory to the recent ESPAC-1 trial results. It is clear from our data that radiation therapy as a potential component of adjuvant treatment, either in the standard or trial setting, cannot be dismissed.