Pancreatic neuroendocrine tumors (PanNETs) are rare, accounting for only 1–2% of all pancreatic neoplasms, with approximately 1000 new cases diagnosed annually in the United States.1,2 These tumors are typically indolent in nature, can be hormonally functional or nonfunctional, and can have low or high malignant potential. Many are discovered incidentally on cross-sectional imaging for other diagnoses.3 Heterogeneity in the clinical presentation of PanNETs creates unique challenges for their management, particularly in deciding on the extent of surgical resection or on surgical resection versus surveillance.4,5

The World Health Organization (WHO) classifies pancreatic neuroendocrine lesions into two main categories: well-differentiated neuroendocrine tumors (NETs) and poorly differentiated neuroendocrine carcinomas (NECs).4,6 Grading of NETs is based on mitotic count and the Ki-67 index.4,6 The Ki-67 index stratifies PanNETs into low-grade tumors (Ki-67 < 3%), intermediate-grade tumors (Ki-67 3–20%), and high-grade carcinomas (Ki-67 > 20%).1,7 Findings have shown the Ki-67 index, in particular, to be highly correlated with clinical outcome, perhaps more so than other known histopathologic features.8,9,10,11

Approximately 90% of diagnosed PanNETs are considered well-differentiated tumors (low-to-intermediate grade) and primarily treated with surgical resection.2,4,12 However, the clinical behavior of these well-differentiated tumors is heterogeneous, with recurrences reported for 10–54% of patients, and the median overall survival (OS) ranges from 51 to 79 months.3,12,13,14 With the majority of resected PanNETs (80%) well-differentiated, and with the Ki-67 index lower than 3% all grouped together, the current WHO stratification scheme is limited in its ability to differentiate outcomes within this low-grade PanNET subset.13,15,16 Thus, our study aimed to further risk-stratify patients with well-differentiated, low-grade PanNETs (Ki-67 index < 3%), with the intent of using Ki-67 to discriminate outcomes better by predicting disease recurrence and biologic behavior.

Methods

Study Population and Pathologic Assessment

All patients with primary, non-metastatic PanNETs who underwent surgical resection at Emory University from 1 January 2000 to 31 December 2013 were identified. Pathologic re-review was undertaken. Only patients who had surgery with curative intent and with available tissue samples for pathologic re-review were included in the study. For the purpose of this study, only patients with well-differentiated tumors and a Ki-67 lower than 3% (low grade) were included in the analysis. Patients with distant metastases or other malignancies, multifocal disease, R2 resections, or mortality within 30 days after surgery were excluded from the analysis.

For this study, tissue microarrays were created in triplicate from the archived formalin-fixed, paraffin-embedded archived tissue blocks. Standard immunohistochemistry was performed. The slides were stained for Ki-67 (clone MIB-1; DAKO Agilent Pathology Solutions, Santa Clara, CA, USA) using antigen retrieval and the Leica Bond Autostainer (Leica Biosystems, Wetzlar, Germany) and counterstained with hematoxylin. The tissue microarray slides were scanned at × 40 magnification on the Leica Aperio AT2 bright field instrument (Leica Biosystems) for computerized quantitation image analysis. The Ki-67 proliferation index was determined as the percentage of tumor cells with Ki-67 immunoreactive nuclei. Depending on the tumor cellularity, the number of total cells counted ranged from 600 to 1800. An overall score for each sample was calculated as the sum of the expression percentage of each triplicate divided by 3. An experienced and dedicated gastrointestinal pathologist at Emory University, blinded to all other clinicopathologic variables for each tissue sample, supervised the analysis. Only the results from re-review of Ki-67 expression were used for analysis.

Study Variables

Retrospective chart review captured pertinent demographic, preoperative, intraoperative, pathologic, and postoperative data. Preoperative comorbidities were defined using the Charlson Comorbidity Scoring System, and staging was assigned as per the American Joint Committee on Cancer 7th edition guidelines.5 Data regarding neoadjuvant and adjuvant therapy, disease recurrence, and survival were additionally collected. Recurrence of disease was specifically determined according to review of patient medical records, radiographic reports on surveillance imaging, and/or biopsy results. Institutional review board approval was obtained before data collection.

Statistical Analyses

Descriptive and comparative analyses were performed on the entire cohort. Recurrence-free survival (RFS) was calculated from the date of operation to the date of recurrence diagnosis. All statistical analyses were performed using SPSS version 23.0 (Armonk New York Software, IBM Inc., Armonk, New York, USA). Chi square analyses and Fisher’s exact tests were used to compare categorical variables, and Student’s t test was used for continuous variables, where indicated. Uni- and multivariable Cox regression analyses were performed to assess the association of individual pathologic factors with RFS. Kaplan–Meier survival plots for RFS were created, and variables were compared using log-rank tests. Statistical significance was predefined as a p value lower than 0.05.

Results

Patient Variables

Of 265 patients with gastrointestinal/pancreatic neuroendocrine tumors and tissue available for pathologic re-review, 98 (37%) had primary PanNETs, and 83 (31%) had well-differentiated, non-metastatic pancreatic tumors that underwent curative-intent resection. After exclusion of 11 patients with a Ki-67 index of 3% or higher, 72 (27%) patients who had primary resected PanNETs with a Ki-67 index lower than 3% were available for analysis.

The baseline demographics and clinicopathologic features of this study cohort are summarized in Table 1. The mean age was 55 years. Nearly half of the patients (46%, n = 33) were male, and 74% (n = 53) were white. The mean tumor size was 3.2 cm, and 26% (n = 19) of the PanNETs were considered functional tumors. The location for 51% of the tumors (n = 37) was in the body or tail of the pancreas, and 71% (n = 51) were resected via an open procedure. Whereas 64 (89%) underwent R0 resections, only 8 (11%) underwent R1 resections. No R2 resections were performed in our study cohort, and 90% of all recurrences were at distant sites. Only 5 of the 72 patients received systemic therapy, 3 of which were treated neoadjuvantly.

Table 1 Baseline demographics and clinicopathologic variables of patients with primary pancreatic neuroendocrine tumors who underwent curative-intent resection at a single institution from 2000 to 2013 (n = 72)
Full size table

Pathologic Re-review and Ki-67 Data

Among the 72 well-differentiated PanNETs with a Ki-67 index lower than 3%, Ki-67 was further stratified into three initial groups after pathologic re-review: group A (< 1%: n = 43, 60%), group B (1–1.99%: n = 23, 32%), and group C (2–2.99%: n = 6, 8%) (Table 2). Representative immunohistochemical KI-67 slides for this grouping are shown in Fig. 1. These groups were well-matched in terms of baseline demographic and operative variables including age, race, comorbidities, location of PanNET, tumor size, and operative blood loss (p > 0.05). However, the groups displayed key pathologic differences. Groups B and C were characterized by advanced T stage (44 and 67%, respectively) compared with group A (12%) (p = 0.003), as well as by an increased incidence of lymphovascular invasion (LVI) (83% in group C vs 23% in Group A) (p = 0.007) (Table 2).

Table 2 Distribution of covariates among patients with low-grade non-metastatic primary pancreatic neuroendocrine tumors who underwent curative-intent resection from 2000 to 2013 at a single institution, stratified by the Ki-67 index
Full size table
Fig. 1
figure 1

Re-stratification of Ki-67 based on histologic analysis by a pathologist blinded to all other clinicopathologic variables using tissue microarray blocks made in triplicate. Patients with low-grade, Ki-67 < 3% were further stratified into 3 initial groups based on immunohistochemical expression: Group A (Ki-67 < 1%), Group B (Ki-67 1–1.99%), and Group C (Ki-67 2–2.99%)

Full size image

Recurrence-Free Survival Analysis

The median follow-up period was 39 months (interquartile range [IQR] 7.1–60.3). In the Kaplan-Meir analysis, the 3-year RFS was 97% for group A, 71% for group B, and 67% for C (p = 0.018) (Fig. 2a).

Fig. 2
figure 2

Kaplan-Meier survival curve for recurrence-free survival in re-stratified Ki-67 index groups. a There is no statistically significant difference between Group B (Ki-67 1–1.99% (n = 23)) and Group C (Ki-67 2–2.99% (n = 6)), but both are associated with worse RFS compared to Group A (Ki-67 < 1% (n = 43)). Log-rank p value = 0.018; b The combined Groups B and C (Ki-67 1–2.99% (n = 29)) show a 27% decrease in 3-year RFS compared with Group A (Ki-67 < 1% (n = 43)). Log-rank p value = 0.005. c After excluding functional and syndromic PanNETs (n = 20), the combined Groups B and C (Ki-67 1–2.99% (n = 20)) show a 23% decrease in 3-year RFS compared with Group A (Ki-67 < 1% (n = 32)). Log-rank p value = 0.017

Full size image

Ki-67 Index as a Guide for Re-stratification

Given the similarity in pathologic characteristics and RFS between groups B and C, these groups were combined to form two final subsets of patients with well-differentiated, low-grade PanNETs for subsequent survival analysis: group A with a Ki-67 lower than 1% (n = 43, 60%) and groups B and C with a Ki-67 of 1–2.99% (n = 29, 40%). Analysis showed that group A had a 3-year RFS of 97%, whereas groups B and C had a decreased 3-year RFS of 70% (p = 0.005) (Fig. 2b).

A subgroup analysis also was performed excluding the 20 patients of the study group with functional and syndromic PanNETs. The results indicated a 23% decrease in 3-year RFS (96% vs 73%) for the combined groups B and C compared with group A (p = 0.017) (Fig. 2c).

Predictors for Recurrence-Free Survival

In the univariable Cox regression analysis, a Ki-67 index of 1% to 2.99% (HR 7.1; 95% CI 1.5–33.8; p = 0.014), tumor size, final resection status, perineural invasion (PNI), lymph node positivity, and advanced T stage each was associated with worse RFS (Table 3). In the multivariable analysis, R1 resection and Ki-67 (HR 8.6; 95% CI 1.0–70.7; p = 0.045) remained independently significant, even when these other adverse clinicopathologic factors were taken into account (Table 3).

Table 3 Association of clinicopathologic factors with recurrence of disease in patients with non-metastatic primary pancreatic neuroendocrine tumors who underwent curative-intent resection from 2000 to 2013 at a single institution
Full size table

Discussion

Although well-differentiated, low-grade PanNETs are uniformly noted to have improved recurrence and overall survival compared with high-grade PanNETs (i.e., pancreatic NECs), significant heterogeneity in outcomes among these well-differentiated tumors remains.12,13 The absence of a reliable risk stratification scheme to discriminate among well-differentiated, low-grade PanNETs has hindered ability to predict disease recurrence adequately after surgical resection.13,17 Furthermore, grouping up to 80% of resected PanNETs as well-differentiated (grade 1 or 2) limits ability to guide individualized postoperative management and surveillance.2,13,18,19

Findings have shown the Ki-67 index to be a particularly sensitive histopathologic marker for a more aggressive clinical course.9,20 However, the current method of using broad Ki-67 categories to define PanNETs may obscure the true prognostic value of this variable, particularly for the low-grade cohort.10 Indeed, our findings suggest that even among well-differentiated, low-grade PanNETs, a Ki-67 index of 1–2.99% is independently associated with an eightfold increase in risk of disease recurrence compared with tumors that have a Ki-67 index lower than 1%, even after the study has accounted for other adverse clinicopathologic variables.

Previous studies have attempted to determine the optimal Ki-67 index cutoff point to predict outcomes for low-to-intermediate grade PanNETs.21 For example, Lowe et al.9 demonstrated that a Ki-67 index above 10% rather than above 3% better predicts lymph node metastases and poor overall survival (OS). Another study by Hamilton et al.22 found that a Ki-67 index higher than 9% predicts a higher likelihood of disease recurrence and worse OS. Finally, a Ki-67 index cutoff of 7.5% was the recommendation of Goodell et al.23 However, these studies included both low- and intermediate-grade PanNETs in their analyses instead of focusing solely on the well-differentiated, low-grade subset. Because our findings suggest heterogeneity in disease recurrence when the Ki-67 index is used for stratification, even for such low-grade PanNETs, it may be inappropriate to increase the Ki-67 cutoff to classify more patients into the low-grade group. Partly due to the rarity of PanNETs, studies that focus on this subset with a Ki-67 lower than 3% currently are limited.

In our study, the PanNET patients with a Ki-67 lower than 1% had improved RFS compared with the patients who had a Ki-67 index of 1–2.99%, with an RFS of 97% at 3 years versus 70%, respectively. Although no studies to our knowledge have focused on the effect of the Ki-67 index on RFS in well-differentiated, low-grade PanNETs, a study by Boyar Cetinkaya et al.3 did show a significant difference in 5-year OS for patients who had PanNETs with a Ki-67 lower than 2% (75.2%) versus a Ki-67 of 3–20% (55.8%) (p = 0.04).3This difference continued in the analysis of 10-year OS in this study, with a Ki-67 lower than 2% related to a 68.9% survival compared with a 46.5% survival for a Ki-67 of 3–20% (p = 0.03).

Another study by Miller et al.19 also analyzed OS for patients whose NETs were graded on the basis of the WHO/European Neuroendocrine Tumor Society (ENETS) stratification scheme. In this study, those classified as having low-grade tumors had an OS of 87% compared with 83 and 50% for the patients with intermediate- and high-grade tumors, respectively.19 However, this study did not focus solely on PanNETs, but rather on all gastrointestinal NETs.

Finally, a study by Panzuto et al.24 also identified the Ki-67 index as a risk factor for disease progression in PanNETs. In their study, compared with low-grade PanNETs, intermediate-grade (Ki-67 3–20%) and high-grade (Ki-67 > 20%) PanNETs were associated respectively with 3.43-fold (p < 0.001) and 1.5-fold (p = 0.074) increases in risk for progression (Ki-67 ≤ 2%).24

To our knowledge, this study is the first to evaluate the use of the Ki-67 index to further risk stratify and predict outcomes for well-differentiated, low-grade PanNETs. These findings may be used to guide future management of low-grade PanNETs given the unclear current National Comprehensive Cancer Network (NCCN) guidelines regarding surgical resection versus observation for small well-differentiated, low-grade PanNETs.15 These results could perhaps be extrapolated to support surgical resection of PanNETs 2 cm in size or smaller if the measured Ki-67 index is higher than 1% on a preoperative biopsy because tumors with a Ki-67 index higher than 1% may display more aggressive behavior. Specifically, this study demonstrated higher recurrence rates after resection for such tumors compared with those smaller than 1%. Moreover, postoperative surveillance could be increased to include imaging at shorter intervals rather than intervals of 1 year or longer, as per the current standard of care.15 Ultimately, this proposed stratification scheme could be included in future adjuvant trials for the development of targeted postoperative therapy, as the armamentarium for PanNETs continues to improve.

This study had various limitations, including its retrospective design using data from a single institution with a relatively small number of patients who had tissue available for analysis. Pathologic re-review was performed only for Ki-67, which left data on other important histopathologic factors such as lymphovascular and perineural invasion dependent on the original surgical pathology reports, from which they were frequently missing. Moreover, the creation of tissue microarrays from random areas in each tumor may have underestimated the Ki-67 index because hot-spot analysis generally is used in the clinical setting. The applicability of our findings is further limited by the numerous and unstandardized methods that currently exist for measuring Ki-67 proliferation and their variable reported accuracy and reliability.23,25

In conclusion, our data suggest that even within well-differentiated, low-grade PanNETs, the Ki-67 index is a strong predictor of disease recurrence. This may have implications for its incorporation into future grading systems, treatment recommendations, and surveillance protocols as the management for PanNETs continues to evolve.