Disease-free survival curve (Kaplan-Meier) for the 181 patients with nonfunctioning pancreatic neuroendocrine tumors (NF-PanNETs) who underwent resection with curative intent.
Disease-free survival curves (Kaplan-Meier) comparing 126 patients with N0 with 55 patients with N1 (P < .001, log-rank comparison).
Receiver operator characteristic (ROC) curve for the first predictive model (A) and second predictive model (B).
eTable. Risk of lymph node metastases according to the proposed predictive models probability
Partelli S, Gaujoux S, Boninsegna L, Cherif R, Crippa S, Couvelard A, Scarpa A, Ruszniewski P, Sauvanet A, Falconi M. Pattern and Clinical Predictors of Lymph Node Involvement in Nonfunctioning Pancreatic Neuroendocrine Tumors (NF-PanNETs). JAMA Surg. 2013;148(10):932-939. doi:10.1001/jamasurg.2013.3376
Nonfunctioning pancreatic neuroendocrine tumors (NF-PanNETs) are often indolent neoplasms without lymph node (LN) metastasis at diagnosis. Therefore, in patients with low risk of LN metastasis, the extent of surgery and lymphadenectomy could be limited and follow-up adjusted to the very low risk of relapse.
To construct a predicting model to assess the risk of pN+ prior to surgical resection for NF-PanNETs using preoperative retrievable variables.
Retrospective review using multiple logistic regression analysis to construct predictive model of pN+ based on preoperatively available data.
The combined prospective databases of the Surgical Departments of the University of Verona, Verona, Italy, and Beaujon Hospital, Clichy, France, were queried for clinical and pathological data.
All patients with resected (R0 or R1), pathologically confirmed NF-PanNETs between January 1, 1993 and December 31, 2009.
Main Outcome and Measure
Risk of lymph node metastases in patients with pancreatic neuroendocrine tumors.
Among 181 patients, nodal metastases were reported in 55 patients (30%) and were associated with decreased 5-year disease-free survival (70% vs 97%, P < .001). Multivariable analysis showed that independent factors associated with nodal metastasis were radiological nodal status (rN) (odds ratio [OR], 5.58; P < .001) and tumor grade (NET-G2 vs NET-G1: OR, 4.87; P < .001) (first model). When the tumor grade was excluded, rN (OR, 4.73; P = .001) and radiological tumor size larger than 4 cm (OR, 2.67; P = .03) were independent predictors of nodal metastasis (second model). The area under the receiver operating characteristic curve for the first and second models were 80% and 74%, respectively.
Conclusions and Relevance
Patients with NF-PanNET-G1 have a very low risk of pN+ in the absence of radiological signs of node involvement. When preoperative grading assessment is not achieved, the radiological size of the lesion is a powerful alternative predictor of pN+. The risk of pathological nodal involvement in patients with NF-PanNETs can be accurately estimated by a clinical predictive model.
Nonfunctioning pancreatic neuroendocrine tumors (NF-PanNETs) represent a rare group of neoplasms with a wide range of aggressiveness.1- 3 A multidisciplinary approach to this disease is mandatory considering patients’ characteristics and tumor natural history. In this setting, surgical resection represents the first-line therapy for tumors limited to the pancreas,4- 6 but extent of surgery should be tailored according to risk-benefit balance. Surgical approaches for NF-PanNETs include standard or typical resections (ie, pancreaticoduodenectomy [PD], distal pancreatectomy [DP], or total pancreatectomy) and atypical or parenchyma-preserving resections (ie, middle pancreatectomy [MP] or enucleation).7 Middle pancreatectomy or enucleation dramatically reduce the risk of long-term pancreatic insufficiency although they are associated with a significant higher risk of postoperative morbidity.8,9 Moreover, when the resection is limited to a single lesion (ie, enucleation) a regional lymphadenectomy is not routinely performed.
The impact on survival of LN metastases in NF-PanNETs has been clearly demonstrated,10- 15 and the 2 proposed TNM classification systems consider pN+ status as a prognostic factor.16,17 However, only about 40% of patients with NF-PanNETs present with LN metastases at diagnosis, suggesting that most patients could be spared a lypmphadenectomy. Despite major improvements in cross-sectional and nuclear imaging techniques, an accurate preoperative staging of nodal status is still challenging.18,19 The identification of preoperative variables associated with the risk of pN+ tumors would be desirable to plan the most appropriate surgical treatment. The objective of this study was to construct a predicting model to assess the risk of pN+ prior to surgical resection for NF-PanNETs using imaging information.
Approval from ethical committees was not required for this retrospective observational study. The current Italian legislation does not require any formal approval for retrospective studies, only a notification to the institution's ethical committee once the study has been made. All the patients signed an informed consent for the use of personal data.
The combined databases of the University of Verona, Verona, Italy, and Hopital Beaujon, Clichy, France, were analyzed to identify patients with NF-PanNETs who underwent surgical resection with curative intent between January 1, 1993 and December 31, 2009. We excluded patients: (1) with functioning tumors, (2) with R2 resections, (3) with distant metastases, (4) who underwent enucleation without lymphadenectomy, and (5) with poorly differentiated neuroendocrine carcinoma (NEC-G3), according to recent recommendations by Kulke et al.20 Information about clinical presentation, demographics, data regarding surgical procedures, postoperative course and complications, pathologic findings, and follow-up was collected.
T staging was performed by conventional imaging techniques including high-resolution computed tomography (CT) and magnetic resonance imaging (MRI), and from 1998 onward, somatostatin receptor scintigraphy (Octreoscan; Mallinckrodt, Inc). In the current study radiological NM staging was assessed by CT. Radiological nodal involvement was reported as suspected in case of enlargement exceeding 1 cm, hypervascularization of peripancreatic lymph nodes, or both.21
According to the anatomical location of the neoplasm, PD or DP with splenectomy were performed, always associated with regional lymphadenectomy. Middle pancreatectomy was performed for tumors located in the body of the pancreas, when feasible, and lymphadenectomy was limited to the peripancreatic nodes. In all the cases, sampling of distant nodes was performed, if metastatic disease was suspected (ie, enlarged lymph nodes). For a PanNET located in the pancreatic head/neck, regional nodes included those located along the common bile duct, common hepatic artery, portal vein, posterior and anterior pancreaticoduodenal arcades, and along the superior mesenteric vein and right lateral wall of the superior mesenteric artery. For a PanNET located in the pancreatic body/tail, regional nodes included those along the common hepatic artery, celiac axis, splenic artery, and splenic hilum. All the patients with metastases in extraregional nodes were considered M1 and then excluded from the present study.
Quality of resection was determined according to the R-classification by the Union for International Cancer Control.22 All the cases were reviewed and classified according to the latest World Health Organization 2010 (WHO-2010) classification23 and were assigned to a TNM-based stage and grading score according to Rindi et al.16 Tumor (T) status was also determined according to Union for International Cancer Control (7th edition).24 The Ki67 proliferative index was expressed as a percentage based on the count of Ki67-positive cells in 2000 tumors cells in areas of the highest immunostaining using the MIB1 antibody (DBA and Dako) and were stratified into 2 categories, G1 (Ki67 < 2%) and G2 (Ki67 ≥ 2% and <20%), according to the 2010 World Health Organization classification.23
Distribution of continuous variables is reported as median and interquartile range (IQR) (25th through 75th percentiles). Categorical variables are presented as numbers and percentages. The comparison between subgroups was carried out using the t test or Mann-Whitney test for continuous variables. Qualitative data were compared by the χ2 test or Fisher exact test when necessary. Disease-free survival (DFS) was defined as the time from resection to disease recurrence and was censored at the last follow-up date if no events had occurred. Survival probability was estimated according to the Kaplan-Meier method. Univariate and multivariate analyses was performed using the Cox proportional hazards regression model to evaluate significant recurrence predictors and their relative role in the cohort. The logistic regression model was used to estimate a clinical predictive model for positive nodes using preoperative retrievable variables. The predictive power of the model was assessed by calculating the area under the receiver operator characteristic curve. All tests were 2-sided. Statistical analyses were performed in SPSS 16.0 for Windows software (SPSS Inc). Statistical significance was set at P ≤ .05.
Between January 1993 and December 2009, 181 patients underwent pancreatic resection with curative intent for NF-PanNETs. Overall, 74 patients (41%) underwent PD, 75 patients (41%) underwent DP, and 32 patients (18%) underwent an MP. Table 1 and Table 2 lists the median number of LNs resected according to different surgical procedures and according to grading,25 respectively. Patients with LN metastases had a higher median number of LN examined as given in Table 3. Postoperative mortality was nil. Half the patients (n = 90) had postoperative complications, including 66 patients (36.5%) with a pancreatic fistula. Patients with postoperative complications had a higher median number of LNs resected (15 vs 12.5, P = .02). The reoperation rate was 4% (n = 8). The median radiological diameter of the primary neoplasm was 29 mm (IQR, 19-44.5 mm) compared with a median histological diameter of 18 mm (IQR, 7-35 mm) (P < .001, Wilcoxon signed rank test). Pathological examination revealed the presence of microscopic involvement of resection margin (R1) in 10 patients (5.5%). Among the 126 patients classified as having pN0, 115 patients (91%) were also classified as having rN0, whereas among the 55 patients classified as having pN1 only 19 patients (35%) were also classified as having rN1. The sensitivity and specificity for CT in detecting LN metastases were then 35% and 91%, respectively. At a median follow-up of 55 months (IQR, 24-84 months), 9 patients (5%) died of disease, whereas another 6 patients (3%) died of unrelated causes. Overall, 23 patients (13%) had a recurrence after surgery. The 1-, 2-, and 5-year DFS rates were 96%, 95%, and 89%, respectively (Figure 1).
In the entire cohort, 55 patients (30%) had LN metastasis. Univariate analysis revealed that age older than 60 years, R1 resection, microangioinvasion, perineural invasion, liver metastases, tumor grade (NET-G2), and LN metastases were associated with a poorer DFS (Table 4). These factors were entered in multivariable analysis, which confirmed that the presence of an NF-PanNET-G2 tumor (hazard ratio [HR], 7.81; P < .001) and nodal involvement (5.21; P = .001) were independent predictors of DFS (Table 4). Particularly, patients with pN+ had a 1-, 3-, and 5-year DFS of 92%, 86%, and 70%, respectively, compared with 99%, 99%, and 97% of patients with N0 (P < .001) (Figure 2). Among 55 node-positive patients, 29 (53%) had more than 1 positive LN. These patients had DFS rates similar to those who had only 1 metastatic LN (5-year DFS: 63% vs 65%, P = .65). Similar results were also found using different cutoff points as 2 (5-year DFS: 66% vs 74%, P = .27) and 5 (5-year DFS: 67% vs 86%, P = .38) metastatic nodes. Among patients with pN+ who had a recurrence (n = 17), 4 patients (7%) eventually died of disease. The site of recurrence in the pN+ group included the liver in most cases (10 patients [59%]) whereas the remaining 7 patients (41%) had local recurrences. Among patients with pN0, 3 different cutoff points of LN number were defined at the 25th, 50th, and 75th percentiles. The log-rank comparison survival analysis failed to demonstrate any survival benefit for patients with pN0 who had more than 7, 14, or 22 LNs evaluated at final histological examination. Among node-negative patients who underwent PD, the number of LN resected did not correlate with DFS (HR, 1.02; 95% CI, 0.96-1.05; P = .92), whereas tumor grade (NET-G2 vs NET-G1) remained the only independent predictor of poor survival in this subgroup (25; 3.71-167.43; < .001).
By univariate analysis (Table 5), factors associated with LN metastasis were maximum radiological diameter larger than 4 cm of primary neoplasm (odds ratio [OR], 2.64; 95% CI, 1.32-5.29; P = .006), tumor site in the pancreatic head (2.68; 1.40-5.13; .003), radiological nodal involvement (rN+) (5.52; 2.40-12.67; P < .001), and tumor grade (G2 vs G1: 5.13; 2.59-10.13; < .001). Multivariable analysis, adjusted for type of pancreatic resection, suggested that the significant independent factors associated with pN+ were rN+ (OR, 5.58; 95% CI, 2.15-14.48; < .001) and tumor grade (G2 vs G1: 4.87; 2.29-10.34; < .001) (first model) (Table 5). If we only considered preoperative available data and consequently removed grading from our model, radiological diameter larger than 4 cm resulted as independent predictor of pN+ (OR, 2.67; 95% CI, 1.12-6.36; P = .03) along with rN+ (4.73; 1.86-12.06; .001) (second model). The area under the receiver operator characteristic curve for the first and second models were 0.80 and 0.74, respectively (Figures 3A and B). The 181 patients were then divided into 2 groups: those with a pN+ risk of 25% or less and those with a pN+ risk greater than 25%. Then, the number of metastatic LNs was tabulated for the 2 groups according to the risk predicted in the 2 models. The results demonstrated that 85% and 87.5% of patients with a pN+ risk of 25% or less were actually N0 in the first and second model, respectively (eTable in the Supplement).
In the absence of suspected LN metastases on preoperative imaging, the presence of an NF-PanNET-G1 is associated with a very low risk of pN+. As a consequence, the presence of these variables is useful for choosing the optimal surgical approach (typical resection vs parenchyma-sparing resection). Because grading could be difficult to assess in the preoperative setting, the radiological diameter is a valid alternative to predict LN involvement.
This large series represents a 15-year experience of 2 referral centers in Europe in the surgical treatment of NF-PanNETs. Nonfunctioning forms represent the vast majority of the neuroendocrine tumors. The DFS analysis showed that both LN metastases and G2 grade are powerful prognostic factors. The prognostic value of pN+ has been demonstrated by several studies,10- 15 and the present series corroborates these results, as patients with LN involvement had a significant higher risk of recurrence compared with those without LN metastases. In the current study, we failed to find an optimal cutoff number of resected LNs associated with improved outcome after resection of an NF-PanNET. Patients with metastatic nodes had a higher median number of resected nodes than patients with pN0. Consequently, several patients with pN0 could have experienced a stage migration effect, thus biasing the results and limiting the value of the recurrence analysis. Nevertheless, and despite this limitation, neither the median number of resected LNs nor the LNR was found to be associated with the risk of recurrence. This result seems to attenuate the weight of the stage migration effect in these tumors.
In patients with NF-PanNETs, if preoperative staging by high-resolution imaging techniques suggests nodal metastases, standard pancreatic resections are commonly performed. On the contrary, a parenchyma-preserving resection could be advocated when the risk of LN metastases is very low. Therefore, a clinical prediction of the LN status becomes of paramount importance for planning the optimal strategy for each patient. In this cohort, we demonstrated that the radiological nodal involvement (rN1) and tumor grade were independent predictors of LN metastases in NF-PanNETs. The presence of one or both of these variables should prompt the surgeon to perform lymphadenectomy avoiding parenchyma-sparing procedures. Because patients with distant metastases in the current study were excluded, firm recommendations about lymphadenectomy, in those patients with resectable liver metastases cannot be made. In this setting, the risk of positive nodes in these patients cannot be assessed by the current predictive model and the value of lymphadenectomy in these patients is unknown. Particularly, patients with a NF-PanNET-G2 are more likely to have LN metastases and, therefore, they presumably benefit from an adequate node clearance. However, the possibility of a preoperative evaluation of Ki-67 is still a matter of debate. To date, fine-needle aspiration cytology represents the only option to estimate Ki-67 value preoperatively. Unfortunately, there is a lack of data about the accuracy of fine-needle aspiration cytology in the assessment of Ki-67 value. Piani et al25 found that Ki-67 expression measured on cytological samples after endoscopic ultrasound has a good agreement with Ki-67 expression on histological sections. In this pilot study, the Ki-67 value was concordant in 89% and 78% of cases for Ki67 values of 2% and of 2% to 10%, respectively. Another limitation of this approach is related to intratumoral Ki67 heterogeneity and to different expression of Ki-67 in primary tumors and metastases. In the future, nuclear medicine assessment, including fludeoxyglucose F 18–positron emission tomography, may help to identify tumors with more aggressive behavior, that is , G2- and/or LN-positive tumors.26 Moreover, the low sensitivity of CT in detecting LN metastases does not allow recommending an atypical pancreatic resection in all patients with rN0. On the other hand, the high specificity of CT suggests to plan lymphadenectomy in patients with rN+.
Considering the limitations of preoperative tumor grade determination, we chose to consider the radiological diameter as an alternative predictor in the model. Indeed, a radiological diameter larger than 4 cm resulted in a powerful association with the risk of LN metastases. Franko et al27 showed that tumor diameter could predict the risk of nodal metastases, and a strict correlation between tumor diameter and poor survival after resection of NF-PanNETs was previously demonstrated.11,15,28 Finally, NF-PanNETs smaller than 2 cm are unlikely to show aggressive behavior, especially in asymptomatic patients.29 In the setting of small tumors, a parenchyma-sparing resection could be safely performed.30
In the current study, we found that the radiological diameter overestimates the pathological diameter (29 mm vs 18 mm). Indeed, the median radiological diameter is large. This is mainly related to the exclusion of those patients who underwent enucleation who had usually smaller lesions. The 4-cm cutoff was considered the median value of patients in the N0 cohort, although other cutoff points were tested in the univariate model.
Another interesting finding of the current study is that the number of metastatic LNs does not affect DFS. We observed that patients with 1 positive node had similar DFS rates compared with those with more than 1, 2, or 5 positive nodes. These results contrast with those reported for pancreatic ductal adenocarcinoma in which the number of metastatic nodes is a powerful prognostic factor.31 These data are difficult to interpret. However, considering the slow and possibly indolent natural history of NF-PanNETs, the short follow-up of our study might explain the aforementioned differences. In pancreatic ductal adenocarcinoma, a meta-analysis has demonstrated that an extended lymphadenectomy is not associated with improved survival.32 In the present study, the number of metastatic nodes was not associated with the risk of recurrence; therefore, we suggest that an extended lymphadenectomy should be avoided also for patients with a high preoperative risk of positive nodes. The predictive model is not a differentiating tool for the management of the extent of lymphadenectomy, but its real value is to support the use of a pancreatectomy with lymphadenectomy instead of atypical resection for tumors larger than 4 cm and/or in the presence of suspected LN metastases on CT. As a consequence, the proposed model does not lead the surgeon toward precise management of the extent of lymphadenectomy, but it represents a valid tool for the decision of performing a lymphadenectomy in NF-PanNETs. This study has several limitations. In particular, the retrospective nature limits in part the value of these results. Moreover, the low accuracy of CT in detecting preoperatively the node metastases implies that many metastases can be frequently missed on CT.
Patients with NF-PanNET-G1 in the absence of radiological LN involvement have a very low risk of pN1. When a preoperative cytological diagnosis is not achieved, the radiological size of the lesion larger than 4 cm is a powerful alternative predictor of pN+. When a pancreatic resection is planned on the basis of this predictive model, a standard resection with regional lymphadenectomy should be performed if the pN+ risk is high. In opposition, a pancreas-sparing resection with only limited peripancreatic node sampling is reasonable because of the need to preserve the long-term pancreatic function in these patients with indolent tumors. The value of additional preoperative exploratory procedures, such as routine fine-needle aspiration of the primary tumor, and the role of nuclear medicine imaging should be evaluated in future studies.
Accepted for Publication: January 31, 2013.
Corresponding Author: Massimo Falconi, MD, Chirurgia Generale B, Policlinico “GB Rossi,” Piazzale LA Scuro, 37134 Verona, Italy (firstname.lastname@example.org).
Published Online: August 28, 2013. doi:10.1001/jamasurg.2013.3376.
Author Contributions: Drs Sauvanet and Falconi share the senior authorship.
Study concept and design: Partelli, Gaujoux, Boninsegna, Scarpa, Sauvanet, Falconi.
Acquisition of data: Partelli, Gaujoux, Boninsegna, Cherif, Scarpa, Ruszniewski, Sauvanet.
Analysis and interpretation of data: Gaujoux, Boninsegna, Crippa, Couvelard, Scarpa, Ruszniewski.
Drafting of the manuscript: Partelli, Boninsegna, Scarpa, Ruszniewski.
Critical revision of the manuscript for important intellectual content: Partelli, Gaujoux, Cherif, Crippa, Couvelard, Scarpa, Ruszniewski, Falconi.
Statistical analysis: Partelli, Boninsegna.
Administrative, technical, or material support: Gaujoux, Couvelard.
Study supervision: Gaujoux, Crippa, Ruszniewski, Sauvanet, Falconi.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported in part by grant FIRB RBAP10AHJB from the Italian Cancer Genome Project (Drs Scarpa and Falconi).
Previous Presentation: This study was presented in part at the North America Neuroendocrine Tumor Society (NANETS); October 20, 2011; Minneapolis, Minnesota, and at the Ninth Meeting of the European Neuroendocrine Tumor Society (ENETS); March 9, 2012; Copenhagen, Denmark.