Although surgical resection is generally recommended for patients with localized pancreatic neuroendocrine tumors (PNETs), the impact of resection on overall survival is unknown. The authors investigated the survival advantage of pancreatic resection using a national database.
This is a retrospective survival analysis of patients with PNETs from the Surveillance, Epidemiology, and End Results database (1988-2002).
A total of 728 patients with PNETs were identified with a median survival of 43 months using Kaplan-Meier survival methods. Resection of tumor was associated with significantly improved survival compared with those patients who were recommended for but did not undergo resection (114 months vs 35 months; P < .0001). This survival benefit was demonstrated for patients with localized, regional, and metastatic disease. A multivariable Cox proportional hazards model was constructed to assess the overall effect of surgical resection on survival, and demonstrated an adjusted odds ratio of 0.48 (95% confidence interval, 0.35-0.66) compared with those who were recommended for surgery but did not proceed to surgery.
Pancreatic neuroendocrine tumors (PNETs) comprise a unique subset of pancreatic tumors. They encompass approximately 1% to 2% of pancreatic tumors, leading to an overall incidence of 1 to 2 cases per 1,000,000. Autopsy series have reported an incidence of 0.5% to 1.5%.1 PNETs, also referred to as “islet cell tumors,” occur most commonly as sporadic cases; however, they can be associated with hereditary tumor syndromes such as multiple endocrine neoplasia (MEN) type I and von-Hippel-Lindau disease.2
A clinical classification differentiating functional versus nonfunctional pancreatic neuroendocrine tumors has been described.2 Functional tumors cause paraneoplastic symptoms through the overexcretion of gastrointestinal hormones (eg, insulin and gastrin). “Nonfunctioning” PNETs (ranging from 10% to 58%)3-6 do not cause paraneoplastic symptoms and often present as metastatic disease. Both functional and nonfunctional tumors can be subject to a delay in diagnosis secondary to nonspecific symptoms and the relative rarity of the disease. Nonetheless, PNETs are associated with prolonged survival when compared with other pancreatic neoplasms such as ductal adeno-carcinoma.7
The mainstay of treatment for PNETs is surgical resection. Patients who have undergone resection have been reported to have a decreased likelihood of future metastasis and disease-specific death compared with patients who did not undergo resection.8-10 However, particularly for nonfunctional tumors, it remains controversial as to whether resection alters the natural history of PNETs. In the subset of patients who present with metastases, resection of the PNET primary is a subject of debate, and removal of neuroendocrine metastases remains an area of open discussion.11-15
The aim of the current study was to determine the unconfounded effect of surgical resection of PNET primary tumors on overall survival. We analyzed data from the Surveillance, Epidemiology, and End Results registry (SEER), which contains data regarding patient characteristics, tumor histology, disease stage, recommendations for surgery, operative data, and overall survival.
MATERIALS AND METHODS
Cases were queried from the SEER registry, an administrative database managed by the National Cancer Institute that samples approximately 26% of the US population, for the years 1988 through 2003.
All subjects with tumors of the pancreas were identified. This cohort was limited by selecting records with International Classification of Diseases for Oncology (third edition) codes consistent with neuroendocrine tumors: 8050 (islet cell tumor), 8151 (insulinoma), 8152 (glucagonoma), and 8153 (gastrinoma). A coding schema for operations of the pancreas was introduced in 1988, and we limited our investigation accordingly. Patients identified through autopsy and death certificates were excluded.
Operative Data and Staging
Composite operative groups corresponding to partial pancreatectomy/excision, total pancreatectomy, and pancreatoduodenectomy (PD) were created to provide consistent definitions of surgical groups over the entire time period of 1988 through 2003 (Table 1).
Table 1. Procedure Codes Used
NOS indicates not otherwise specified.
Partial pancreatectomy/pancreatic excision
Local or partial surgical excision of pancreas
Local excision of tumor, NOS
Radical regional (partial) pancreatectomy with lymph node dissection and adjacent soft tissue resection
Partial pancreatectomy, NOS
Total pancreatectomy with/without splenectomy
Total pancreatectomy and subtotal gastrectomy or duodenectomy
Subtotal gastrectomy, duodenectomy with complete or partial pancreatectomy with/without splenectomy (Whipple procedure)
Local or partial pancreatectomy and duodenectomy WITHOUT distal/partial gastrectomy “without subtotal gastrectomy” WITH partial gastrectomy (Whipple procedure)
Incisional, needle, or aspiration biopsy of other than primary site
Incisional, needle, or aspiration biopsy of primary site
Exploratory only (no biopsy)
Bypass surgery, -ostomy only (no biopsy)
Exploratory only and incisional, needle, or aspiration biopsy of primary site or other sites
Bypass surgery, -ostomy only, and incisional, needle, or aspiration biopsy of primary site or other sites
Noncancer-directed surgery, NOS
SEER classifies the documented recommendation for operative and nonoperative management. Nonoperative patients can be classified as: 1) not recommended for surgery because of cancer-related contraindications; 2) not recommended because of noncancer contraindications; and 3) recommended but not performed.
The American Joint Commission on Cancer has no unified staging criteria for endocrine tumors of the pancreas.16, 17 In 2000, the World Health Organization introduced a more structured classification of tumors of the endocrine system,18 which relies on detailed histopathology19, 20; this is not reflected in SEER. For our analyses, we used “SEER historic stage,” which is a coding schema with 3 clinically relevant categories: localized, regional, or distant disease.
Univariate analyses comparing patient demographics and tumor characteristics for each operative group were performed. Unadjusted survival was compared using Kaplan and Meier methods.21
To achieve a “like-to-like” survival comparison of patients who underwent pancreatic resection with those who were recommended for but did not undergo resection, SEER data regarding recommendation for cancer-directed surgery were used in conjunction with SEER data on actual procedures. Thus, patients were divided into Resected, Recommended/Not Resected, and Not Recommended groups.
To ensure that patients who were recommended for but did not undergo pancreatic resection were not simply patients who had unsuccessful attempts, thereby potentially making intergroup comparisons less valid, Recommended/Not Resected patients were classified into 3 subsets based on whether any surgery was performed. A subset of patients who underwent nonresection procedures, but had nonetheless been coded in SEER as receiving cancer-directed surgery, was identified (Recommended/Not Resected/Cancer-directed Other Surgery). A second subset undergoing nonresection procedures that were coded as non–cancer-directed was also identified (Recommended/Not Resected/Non–cancer-directed Other Surgery). The first subset represented biopsies, metastasectomies, et cetera, whereas the latter subset included patients undergoing procedures such as palliative bypasses and feeding tubes. A third subset was comprised of patients recommended for resection who received no surgery of any type (Recommended/No Surgery). For use in a final sensitivity analysis investigating the extreme possibility that all patients who underwent any type of surgery represented attempted resections, patients receiving any type of operative procedure were pooled into 1 group (Recommended/Any Surgery) and compared with Recommended/No Surgery patients.
A Cox proportional hazards model was used to determine the effect of pancreatic resection on survival, adjusting for potential confounders including age, sex, marital status, tumor location, and histologic type. Furthermore, we performed subgroup analyses in which we determined the effect of pancreatic resection on survival within key subgroups adjusted for the aforementioned remaining potential confounders. Effect modification by these variables on the benefit of resection was tested by including formal interaction terms in adjusted Cox models.
All statistical analyses were performed using SAS statistical software (version 9.1; SAS Institute Inc., Cary, NC), with a P value <.05 considered significant.
There were 728 patients with PNETs between 1988 and 2003. One patient whose resection status was unknown was excluded. Demographics are shown in Table 2. The mean age of the patients was 57.2 years (standard deviation, 15.1). Fifty-four percent were men. The majority were white (85.8%); 8.2% were black, and 6.1% were classified as other. No race data were available for 0.6%.
Table 2. Patient Characteristics for All Patients With Pancreatic Neuroendocrine Tumor, 1988-2003
Cancer-directed Surgery Recommended
SEER indicates Surveillance, Epidemiology, and End Results; SD, standard deviation.
One patient was excluded from further analysis (cancer-directed surgery status was unknown).
Islet cell/nonsecreting histology comprised 84.3% (n = 613). There were 43 patients with insulinomas (5.9%), 25 with glucagonomas (3.4%), and 46 with gastrinomas (6.3%). Pancreatic head tumors accounted for 27.9% (n = 203). Pancreatic body tumors were found in 60 patients (8.2%) and tail tumors in 172 (23.8%). One patient had a PNET located in the duct, and 65 patients (8.9%) had multiple foci of tumor within the pancreas. No localizing data were available for 226 patients (31%). SEER stage data were available in 92.3% of patients. Localized disease was found in 13.3% of subjects (n = 97), regional disease in 24.8% (n = 180), and distant disease in 54.2% (n = 394).
Recommendations for Cancer-directed Surgery
Cancer-directed surgery was recommended for 425 (58.5%) patients, 310 of whom (72.8%) underwent pancreatic resection (Resected). The remaining 115 patients who were recommended for cancer-directed surgery did not undergo resection (Recommended/Not Resected). The recommendations for and subsequent receipt of surgery is diagrammed in Figure 1.
Not Recommended patients
Cancer-directed surgery was not recommended for 302 patients (41.5% of entire sample; Not Recommended). Of these, 287 patients (95.0% of Not Recommended) possessed contraindications to surgery stemming from cancer-related factors. Fifteen patients (5.0% of Not Recommended) were deemed to be nonoperative candidates because of comorbidities. One hundred forty patients (46.4% of Not Recommended) underwent nontherapeutic operations. The distribution of patients according to recommendation and resection status is shown in Figure 1.
Operative and Nonoperative Characteristics
Among the 310 pancreatic resections, partial pancreatectomy/enucleation was performed in 158 (51.0%), PD in 112 (36.1%), total pancreatectomy in 26 (8.4%), and nonspecified pancreatectomy in 14 (4.5%).
Patients who underwent resection were younger than nonresected patients (mean age, 54.1 years vs 59.5 years; P < .0001). Women demonstrated a nonsignificant trend toward being resected (46.1% vs 39.7% of men; P = .08). There were no differences in the likelihood of pancreatic resection among races or between functioning and nonfunctioning tumors (P = .40 and P = .13, respectively).
Resected patients by stage
There were significant differences in resection based upon SEER stage; 83.5% of patients with local disease, 72.8% of patients with regional disease, and 22.3% of patients with distant disease underwent resection (P < .0001). Tumors of the tail were more likely to be resected compared with head and body (59.3% vs 42.9% and 46.7%, respectively; P = .006). Furthermore, tumor location affected the type of resection performed (P < .0001). Tumors of the head of the pancreas were significantly more likely to undergo PD (70.1% of pancreatic resections for head lesions were PDs) compared with tumors in the body/tail (19.5% PDs). Of patients with body/tail tumors, 68.2% underwent partial pancreatectomy/enucleation compared with 19.5% of patients with head tumors.
Survival: Univariate Analysis
Survival for all patients
Median survival for all patients with PNETs was 43.0 months (Table 3). The median and mean follow-up were 26 months and 69 months, respectively. Patient race did not influence overall survival (P = .82). Women had longer median survival (47 months vs 38 months; P = .03). Increasing age was associated with worsened survival, with those aged >70 years having a median survival of 21 months, versus a median survival of 76 months in patients aged <50 years (P < .0001).
Table 3. Univariate Kaplan-Meier Survival Analysis of Patient Characteristics
Median Survival, mo
SEER indicates Surveillance, Epidemiology, and End Results; NA, not applicable.
All patients, N=727
Age group, y
Single (never married)
NA mean 101.8
Type of resection
Nonspecified, nonpancreatic surgery
Not recommended (302)
Recommended/Not Resected (115)
Cancer-directed surgery (40)
Noncancer-directed surgery (22)
No surgery (53)
Tumor stage was found to be highly correlated with median survival (P < .0001). The median survival of patients with localized disease could not be calculated; the mean survival was 102 months. The regional disease group had a median survival of 89 months, and the median survival of patients with distant stage disease was 24 months. Unstaged patients survived a median of 26 months.
Survival by resection type
Type of procedure performed was associated with significant differences in median survival. Total pancreatectomy predicted poor prognosis compared with PD and partial pancreatectomy/enucleation (median survival of 52 months vs 123 months and 120 months, respectively; P < .001). Unspecified pancreatectomy patients had a median survival of 110 months. Nonspecified, nonpancreatic surgery patients had a median survival of 22 months.
Comparison of patients recommended for cancer-directed surgery who did/did not undergo resection
Patients undergoing resection had significantly longer overall survival than patients who were recommended for but did not undergo cancer-directed surgery (Resected, 114 months vs Recommended/Not Resected, 35 months; P < .0001) (Fig. 2a). Of the 115 Recommended/Not Resected patients, 40 underwent nonresection procedures coded as cancer-directed (Recommended/Not Resected/Cancer-directed Surgery), and 22 underwent nonresection procedures not coded as cancer-directed (Recommended/Not Resected/Non–cancer-directed surgery). The survival of these 2 subgroups was 48 months and 22 months, respectively. The remaining 53 patients (Recommended/No Surgery) had a median survival of 25 months (Fig. 2b), which remained significantly different from the Resected patients (P < .0001); by way of comparison, the Not Recommended group (n = 302) had a median survival of 21 months.
To approximate an “intent to treat” comparison, a sensitivity analysis was performed as described earlier, using the extreme assumption that all Recommended/Not Resected/Any Surgery patients (n = 372) underwent surgery with the intent to resect; this group had a median survival of 100 months, which remained significantly different from the Recommended/No Surgery group median survival of 25 months (P < .001).
Survival by SEER Stage
Resection was associated with a significant survival benefit in patients in all SEER stage groups (Table 4). For patients with localized PNETs, the median survival for resected patients had not been reached at the time of last follow‒up; however, the mean survival was 111 months. Localized Recommended/Not Resected group had a median survival of 61 months; localized Recommended/No Surgery patients had a median survival of 62 months. However, because of the small number of patients in the localized Recommended/Not Resected group (n = 4), the survival advantage in Resected versus Recommended/Not Resected groups was not statistically significant. Not Recommended localized patients had a median survival of 29 months.
Table 4. Univariate Kaplan-Meier Analysis of Patients in Each Stage by Resection Status
Median Survival, mo
NA indicates not applicable; NS, not significant.
NA (mean 101.8)
NA (mean 111.1)
44 (mean 41.3)
NA (mean 61.0)
Resection vs Recommended/Not Resected, P = NS
29 (mean 32.1)
Resection vs Recommended/Not Resected, P = .009
Distant stage disease
Resection vs Recommended/Not Resected, P = .01
Unknown stage disease
P = .007
Resection vs Recommended/Not Resected, P = .01
P < .001
Resection vs Recommended/Not Resected, P < .0001
For regional disease, the median survival in Resected patients was 129 months, which is significantly different from Recommended/Not Resected patients, whose median survival was 64 months (P = .009). Recommended/No Surgery regional disease patients had a median survival of 42 months; Not Recommended regional-stage patients had a median survival of 49 months.
The survival advantage of pancreatic resection was also demonstrated in patients with distant stage disease. Resected patients with distant stage PNETs demonstrated a median survival of 60 months, which was significantly different from that for Recommended/Not Resected patients, who had a median survival of 31 months (P < .0001). Recommended/No Surgery distant disease patients had a median survival of 22 months; Not Recommended patients had a median survival of 16 months.
A multivariate Cox proportional hazards model was performed to define the factors independently influencing survival (Table 5). Race and sex were not found to be independently predictive of survival (P = .42 and P = .58, respectively). Tumor location and tumor functional status did not significantly appear to affect survival (P = .19 and P = .87). Increasing age per year was predictive of worsened survival (hazards ratio [HR] of 1.03; 95% confidence interval [95% CI], 1.02-1.04 [95% P < .0001]). Regional and distant disease predicted worse survival compared with localized disease (regional HR of 1.72; 95% CI, 1.09-2.74; distant HR, 2.72; 95% CI, 1.74-4.24 [overall P < .0001]). Patients who underwent resection had a survival advantage compared with Recommended/Not Resected patients (HR of 0.48; 95% CI, 0.35-0.66), whereas Not Recommended patients had a survival disadvantage (HR of 1.34; 95% CI, 1.03-1.74 [overall P < .0001]).
Table 5. Multivariate Cox Proportional Hazards Analysis of Survival by Surgical Recommendation Group After Controlling for Patient Covariates and Stratified by Stage
Hazards Ratio (95% Confidence Limit)
NOS indicates not otherwise specified; SEER, Surveillance, Epidemiology, and End Results. Hazards ratios are modeled against risk of death. Age is used as a continuous variable, all others as categoric.
Age (per y)
Nonfunctional (islet cell)
Recommended, not done
Subgroup analyses were performed to determine the effect of pancreatic resection on survival within key subgroups including stage groups (localized, regional, and distant); location groups (head vs nonhead); histology groups (islet cell vs functional); and age groups (<50 years, 50-69 years, >70 years). Including formal interaction terms in adjusted Cox models for each of these variables demonstrated no effect modification on the benefit of resection. The adjusted HRs in the key subgroups are shown in Figure 3.
In the current study, we used the SEER registry to examine patients with PNETs at a national level to determine the true effect of surgical resection.
Patients in our study demonstrated an overall median survival of 44 months. The median survival in patients with localized disease exceeded current follow-up; in patients with regional disease it was in excess of 7 years; and it was approximately 2 years in patients with distant stage disease. The majority of PNETs were diagnosed as distant stage disease (54.3%), rather than regional (24.9%) or localized (13.4%) stages. Of patients in these respective stage groups, 60.7%, 16.7%, and 12.4%, respectively, were recommended for cancer-directed surgery. Most patients who were not recommended for surgery were believed to have cancer-related contraindications, likely from a perceived unresectable primary lesion and/or metastatic disease, although a smaller percentage were documented to have noncancer contraindications.
Approximately 20% of patients who were recommended for surgery did not undergo surgical resection. The sizable proportion of patients who do not undergo recommended surgery becomes more relevant when considered in the context of our data demonstrating that patients with PNETs who undergo surgical procedures of any kind have a marked survival advantage over patients who do not undergo surgery.
Population-based studies of PNETs have generally focused on overall survival based on demographics, tumor stage, and histology rather than by treatment strategy.22, 23 Although a recent study from the National Cancer Database highlighted nationwide disparities in the surgical treatment of PNETs,24 the majority of reports on surgical resection are case series.11, 13, 25-31
Although studies with sufficient sample size with which to compare resected and unresected patients with PNETs are rare, population-based studies in other tumors such as pancreatic adenocarcinoma have demonstrated a survival benefit from surgery.32, 33 However, studies comparing resected versus not-resected patients generally have residual “confounding by indication,” (ie, patients who are recommended for and undergo surgery are different from patients who do not undergo surgery, thus diminishing their validity). In our work, we attempt to control for confounding by indication by comparing resected patients with their peers who were recommended for surgery but did not undergo resection. Strikingly, patients who were recommended for surgery (ie, those who were believed to have neither cancer nor noncancer-related contraindications for surgery) but did not undergo surgery, had a median survival of 28 months, which is similar to the 21-month median survival noted in patients with contraindications to surgery. The survival advantage of resection appears to hold true across patients with local, regional, and distant stage PNETs.
The reasons 17% of the patients who were deemed eligible to undergo surgery did not undergo surgery cannot be gleaned from the SEER registry. Prohibitive comorbidities and patient refusal are coded within the SEER database and thus should not unduly bias our results. Although selection bias and misclassification may still be present, they should not explain the overall findings.
The SEER cancer registry provides a powerful mechanism with which to study demographic and tumor characteristics and associated survival data at a national level. However, there are some limitations to using administrative data. Specific individual characteristics for each record are unknown. Medical comorbidities and patient performance status are not reported, but are extrapolated from the recommendation for or against cancer-directed surgery. Similarly, it is not possible to determine whether the operations were performed for cure or for palliation; R (margin) status and nonresected disease are unavailable. Two areas that warrant future study are the relation between tumor biology (ie, the presence of MEN syndromes) and the relation between different sites of metastatic spread on survival in these patients. Both of these have been shown to be associated with survival in smaller institutional studies; however, these data are not available in the SEER dataset. Finally, SEER has undergone multiple revisions, with different coding schemas. In the current study, we have attempted to control for this by selecting only those cases beyond 1988 when pancreatic surgical procedures were more precisely documented.
In conclusion, patients with PNETs are often diagnosed with advanced disease. A proportion of patients are recommended for surgery; the majority of their nonrecommended peers are excluded for cancer-related factors, including metastatic disease. A smaller subset of patients subsequently undergoes surgery as recommended. The current study data demonstrate a dramatic survival difference between those patients with PNETs recommended for surgery who underwent resection and those who did not. Furthermore, patients who underwent resection had a similarly powerful survival advantage over patients who were recommended for surgery, as over patients with contraindications to cancer-directed surgery. Finally, the survival advantage of surgical resection held true across all SEER stages, included advanced stage PNET disease. If in fact some patients not recommended for surgery are being excluded solely for metastatic disease, our data suggest that the pool of operative candidates should be further expanded. Because surgical resection remains the mainstay of PNET treatment, our data suggest that patients with PNETs may be systematically undertreated.
An often-posed question regarding the available case series in the literature demonstrating favorable outcomes after surgical resection of PNETs is the following: is the lengthy survival associated with resection due merely to biology of the disease rather than the operative intervention? Based on the results of the current study, we believe that the answer is an unequivocal no. Patients with PNETs who undergo surgery demonstrate a distinct survival benefit compared with their unresected peers. Efforts to improve the early diagnosis of PNETs, the careful consideration of cancer-directed surgery in patients with local, regional, and distant disease, and concerted efforts to ensure that patients who are recommended for surgery have the opportunity to undergo surgical resection may help improve overall survival for patients with these unique tumors.
We thank Marc S. Sabatine, MD, MPH, for statistical expertise.
Conflict of Interest Disclosures
Supported by the PanCAN-Samuel Stroum ASCO Young Investigator Award, by a Howard Hughes Early Career Award, and by the Pancreatic Cancer Alliance (all to J.F.T.).