Combined Infiltrative Macroscopic Growth Pattern and Infiltrative Microscopic Tumor Border Status Is a Novel Surrogate Marker of Poor Prognosis in Patients With Pancreatic Neuroendocrine Tumor

After approval with patient consent waiver from the institutional review board (approval number 2014-0580), surgically resected PanNET cases from 2000 to 2015 were selected. Patients with PanNETs who underwent neoadjuvant therapy or PanNET cases with unavailable paraffin-embedded blocks were excluded from this study. We sampled all areas of the tumor with different appearance, and at least one representative section per 1 cm of the tumor was submitted at the time of gross examination.³⁹  In addition, poorly differentiated NECs, including small cell or large cell NECs, were excluded. Finally, a total of 203 PanNET cases were included for histopathologic review. Clinical data, including age, sex, symptoms, procedure, and survival outcomes, were obtained from reviewing electronic medical records.

Pathologic data of PanNETs, including tumor location and size, were retrieved from the surgical pathology reports. Gross examination of the tumor border was reviewed in cases with available gross images of the cut surface of the PanNET. All hematoxylin and eosin–stained slides were carefully reevaluated by 3 pathologists blinded to the clinicopathologic information.

Growth pattern and tumor border were evaluated at the junctions between tumor and normal pancreatic parenchyma or between tumor and other adjacent organs by both gross and microscopic examination. First, macroscopic growth pattern was evaluated in PanNET cases with available gross images of the cut surface of the PanNETs and was classified as expansile or infiltrative pattern. Expansile growth pattern was defined when the tumor had well-delineated margin from adjacent parenchyma (Figure 1, A and B), whereas infiltrative growth pattern was defined when there were any foci of poorly demarcated margin from surrounding structure (Figure 1, C and D). All available gross images were examined. Tumors with even a small focus of macroscopic infiltrating growth pattern were considered to have an infiltrative growth pattern.

Second, microscopic tumor border was categorized as pushing or infiltrative. When the tumor boundary was well circumscribed under microscopic examination, this was referred to as a pushing border (Figure 2, A). PanNETs with smooth, well-circumscribed border by the well-demarcated fibrotic tumor capsule with inclusion of small irregular clusters or cords of infiltrating tumor cells within the fibrotic tumor capsule without accompanying frank tumor cell invasion to adjacent normal pancreatic parenchyma or other organs were also classified as having a pushing border (Figure 2, B). Multinodular tumor with clear delineation of the invasive edge from the adjacent tissue was also considered as pushing border. In contrast, infiltrative border was defined when irregular tonguelike projections of tumor cells or clusters were dissecting through normal pancreatic parenchyma (Figure 2, C) or surrounding structure without a recognizable, distinct delineation of the tumor (Figure 2, D). For better interobserver reproducibility, the tumor border was evaluated with this 2-tier system by assigning tumors with at least focal infiltrative growth pattern to infiltrative border.⁴⁰  All available slides were examined. Areas difficult to interpret were reevaluated via discussion to arrive at a consensus interpretation among the 3 pathologists.

Tumor grade was evaluated by counting mitoses per 10 high-power fields (HPFs) and Ki-67 labeling index according to the current guidelines of the 2017 and 2019 WHO classification systems.^41,42  G1 was assigned to tumors with a mitotic rate of less than 2/10 HPFs and/or a Ki-67 labeling index lower than 3%, G2 to tumors with a mitotic rate of 2 to 20/10 HPFs and/or a Ki-67 labeling index of 3% to 20%, and G3 to tumors with a mitotic rate more than 20/10 HPFs and/or a Ki-67 labeling index higher than 20%.^1,42  Lymphovascular and perineural invasions, marginal status, lymph node, and distant metastatic status were also evaluated. Tumor stage was evaluated based on the 8th edition of the American Joint Committee on Cancer (AJCC) Cancer Staging Manual.⁴³ 

Tissue microarrays (TMAs) were constructed from archived, formalin-fixed, paraffin-embedded tissue blocks with a manual tissue microarrayer (Uni TMA Co Ltd, Seoul, Korea).¹⁵  Three representative cores from areas of the PanNET and one core from normal pancreas parenchyma were punched from the donor block of each PanNET case and transferred to a recipient TMA block to obtain uniform immunohistochemical labeling and normal control, respectively.

Subsequently, 4-μm-thick tissue sections were deparaffinized and hydrated in xylene and serially diluted in ethanol. Endogenous peroxidase was blocked via incubation in 3% H₂O₂ for 10 minutes. Thereafter, heat-induced antigen retrieval was performed. Primary antibodies were used with a Benchmark autostainer (Ventana Medical Systems, Tucson, Arizona) following the manufacturer's protocol. Sections were incubated at room temperature for 32 minutes in primary antibodies for insulin (1:100; 2D11-H5, Novocastra, Newcastle, United Kingdom), somatostatin (1:4000; A0566, DakoCytomation, Glostrup, Denmark), serotonin (1:200; 5HTH209, DakoCytomation), gastrin (prediluted; A0568, DakoCytomation), and glucagon (1:400; 295A-15, Cell Marque, Rocklin, California). The sections were then labeled with an automated immunostaining system with an I-View detection kit (Benchmark XT; Ventana). The immunolabeled sections were lightly counterstained with hematoxylin, dehydrated in ethanol, and cleared in xylene. Immunoreactivity was interpreted by light microscopic examination and independently evaluated by 2 pathologists, who were blinded to the clinicopathologic information. The results were interpreted as positive for diffuse cytoplasmic positivity.

Continuous variables were compared using the independent Student t test or the Mann-Whitney U test. Categorical variables were examined using the χ² test or the Fisher exact test. Concordances of the macroscopic and microscopic patterns between pathologists were evaluated using the Cohen κ statistic.⁴⁴  The κ value was interpreted as follows: 0, no agreement; 0.20 or less, slight agreement; 0.21 to 0.40, fair agreement; 0.41 to 0.60, moderate agreement; 0.61 to 0.80, substantial agreement; and 0.81 to 1, almost perfect agreement.⁴⁴  Survival curves were estimated using the Kaplan-Meier method, and differences between the curves were compared using the log-rank test. The prognostic potential of the factors was evaluated using univariate and multivariate Cox proportional hazards regression analyses. All statistical evaluations were performed using R (version 4.0.3; The R Foundation for Statistical Computing, Vienna, Austria). A P value < .05 was considered to indicate statistical significance.

The mean age of patients was 59.0 ± 12.8 years (median, 60.5 years; range, 35–85 years). There were 96 men (47.3%) and 107 women (52.7%). One hundred sixty-one patients (79.3%) did not display any symptoms, whereas 42 patients (20.7%) had symptoms specific to functioning PanNETs. Forty-eight (23.6%), 41 (20.2%), 5 (2.5%), 5 (2.5%), and 9 cases (4.4%) were immunolabeled to insulin, glucagon, gastrin, serotonin, and somatostatin, respectively. Mean tumor size was 3.0 ± 2.2 cm. When dichotomization was performed, 90 cases (44.3%) were smaller than 2 cm and 113 (55.7%) were 2 cm or larger. There were 130 G1 (64.0%), 61 G2 (30.0%), and 12 G3 (6.0%) PanNETs. Eighty-eight (43.3%) were pT1, 52 (25.6%) were pT2, 57 (28.1%) were pT3, and 6 (3.0%) were pT4 cases based on the T classification of the 8th edition of the AJCC Cancer Staging Manual.⁴³  Thirty (14.7%) and 6 cases (3.0%) showed lymph node metastasis and distant metastasis at the time of diagnosis, respectively. Lymphovascular and perineural invasions were noted in 55 (27.1%) and 30 cases (14.8%), respectively. Eighteen cases (8.9%) had a positive resection margin. The median follow-up period was 71 months (range, 1–137 months).

Macroscopic growth pattern evaluation was performed with 167 cases (82.3%) that had available images of the cut surface of the PanNETs. Of these cases, 83 (49.7%) had an expansile growth pattern whereas 84 (50.3%) displayed an infiltrative pattern to surrounding pancreatic parenchyma or adjacent organs (Figure 1). The associations between growth pattern and clinicopathologic factors are summarized in Table 1. Briefly, PanNETs with grossly infiltrative growth pattern were correlated with tumors that had the following features: nonfunctioning (P < .001), larger size (P = .02), microscopic infiltrative tumor border (P < .001), higher tumor grade (P = .01), higher pT classification (P = .007), lymph node metastasis (P < .001), distant metastasis (P = .04), higher AJCC stage grouping (P < .001), lymphovascular (P = .007) and perineural (P < .001) invasions, and no insulin expression (P = .001).

Microscopic tumor borders were evaluated in the 203 PanNETs. The pushing-type tumor border was observed in 122 cases (60.1%), whereas the infiltrative tumor border was identified in 81 cases (39.9%), based on microscopic examination (Figure 2). The correlation of microscopic tumor border with the clinicopathologic factors of the PanNET cases is summarized in Table 2. PanNETs with infiltrative tumor border were more frequently observed in patients of older age (P = .04) and men (P < .001), and were associated with tumors with the following features: nonfunctioning (P = .03), larger size (P = .003), macroscopic infiltrative growth pattern (P < .001), higher tumor grade (P < .001), higher pT classification (P < .001), lymph node metastasis (P < .001), higher AJCC stage grouping (P < .001), lymphovascular (P < .001) and perineural (P < .001) invasions, involvement of resection margin (P = .007), and no insulin expression (P = .004).

Combined macroscopic growth pattern and microscopic tumor border analysis was performed with 167 cases (82.3%) that had available images of the cut surface of the PanNETs. Based on combined macroscopic growth pattern and microscopic tumor border status, 65 cases (38.9%) were classified as having expansile growth and pushing border (Exp/Pus), 34 (20.4%) had infiltrative growth and pushing border (Inf/Pus), 18 (10.8%) had expansile growth and infiltrative border (Exp/Inf), and 50 (29.9%) had infiltrative growth and infiltrative border (Inf/Inf) (Table 3). PanNETs with Exp/Pus status were more commonly observed in female patients (P = .01) and were associated with tumors of smaller size (<2 cm; P = .01), lower pT classification (T1; P = .004), no lymph node metastasis (P < .001), no distant metastasis (P = .03), lower AJCC stage grouping (P < .001), and insulin expression (P = .001). In contrast, PanNETs with Inf/Inf status were more frequently associated with tumors with no symptoms (P = .01) and lymphovascular and perineural invasions (both P < .001).

We also assessed the 2-tier classification of combined macroscopic growth pattern and microscopic tumor border status after merging Exp/Pus, Inf/Pus, and Exp/Inf into “others” and then compared PanNETs having the “others” pattern with those having Inf/Inf. PanNETs with Inf/Inf status were more frequently observed in male patients (P = .01) and were associated with tumors with no symptoms (P = .01), larger size (≥2 cm; P = .002), higher tumor grade (P < .001), higher pT classification (P < .001), lymph node metastasis (P < .001), distant metastasis (P = .01), higher AJCC stage grouping (P < .001), lymphovascular and perineural invasions (both P < .001), and no insulin expression (P = .007).

Interobserver variability was tested in 100 representative cases by 2 pathologists for tumor border assessment. Interobserver agreement was almost perfect for macroscopic growth pattern (κ = 0.86), moderate for microscopic tumor border (κ = 0.58), and substantial for combined macroscopic and microscopic tumor border (κ = 0.75).

Patients with PanNET having the expansile growth pattern (5-year survival rate, 98.7%) had significantly better overall survival (OS) than those having the infiltrative pattern (76.2%; P < .001; Figure 3, A). Similarly, patients with PanNET having the expansile growth pattern (5-year survival rate, 90.3%) had significantly better recurrence-free survival (RFS) than those having the infiltrative pattern (68.1%; P < .001; Figure 3, B).

Patients with PanNET having the pushing tumor border (5-year survival rate, 90.7%) had significantly better OS than those having the infiltrative border (72.5%; P < .001; Figure 3, C). Similarly, patients with PanNET having the pushing tumor border (5-year survival rate, 83.9%) had significantly better RFS than those having the infiltrative border (62.1%; P < .001; Figure 3, D).

The overall 5-year survival rates in patients with PanNET having Exp/Pus, Inf/Pus, Exp/Inf, and Inf/Inf status were 98.4%, 89.7%, 100%, and 67.2%, respectively (overall comparison, P < .001; Figure 4, A). In the pairwise comparisons, patients with PanNET having Inf/Inf status had significantly worse OS than those having Exp/Pus (P < .001), Inf/Pus (P = .01), and Exp/Inf (P = .04) status. However, no survival difference was observed between patients with PanNET having Exp/Pus and Inf/Pus (P = .32), Exp/Pus and Exp/Inf (P = .53), or Inf/Pus and Exp/Inf status (P = .65). Also, the recurrence-free 5-year survival rates of patients with PanNET having Exp/Pus, Inf/Pus, Exp/Inf, and Inf/Inf status were 92.0%, 82.5%, 83.6%, and 56.7%, respectively (overall comparison, P < .001; Figure 4, B). In the pairwise comparisons, patients with PanNET having Inf/Inf status had significantly worse RFS than those having Exp/Pus (P < .001) and Inf/Pus (P = .004) status, but had a marginally significant difference from patients with Exp/Inf status (P = .05). In contrast, no survival difference was observed between patients with PanNET having Exp/Pus and Inf/Pus (P = .31), Exp/Pus and Exp/Inf (P = .4), or Inf/Pus and Exp/Inf status (P = .98).

Therefore, we merged Exp/Pus, Inf/Pus, and Exp/Inf groups into one group, categorized them as “others,” and compared the OS of patients in this group with that of patients with PanNET having Inf/Inf status. Patients with PanNET having Inf/Inf status (5-year survival rate, 67.2%) had significantly worse OS than those having other growth patterns and/or tumor borders (96.2%; P < .001; Figure 4, C). Also, patients with PanNET having Inf/Inf status (5-year survival rate, 56.7%) had significantly worse RFS than those having other growth patterns and/or tumor borders (88.4%, P < .001; Figure 4, D).

Infiltrative macroscopic growth pattern (hazard ratio [HR], 5.249; 95% CI, 1.794–15.36; P = .002), infiltrative microscopic tumor border (HR, 3.211; 95% CI, 1.655–6.228; P < .001), and Inf/Inf status (HR, 6.094; 95% CI, 2.525–14.7; P < .001) predicted poor OS based on univariate analysis (Table 4). Other factors associated with the OS of patients with PanNETs included age (P < .001), symptoms (P = .02), tumor size (P = .01), tumor G3 (P < .001), pT classification (P < .001), lymph node metastasis (P < .001), distant metastasis (P = .04), lymphovascular and perineural invasions (both P < .001), resection marginal status (P < .001), and glucagon expression (P = .03).

Multivariate analyses were performed to determine whether combined macroscopic growth pattern and microscopic tumor border status would remain a predictor of OS after adjusting for these factors. As T classification, lymph node metastasis, and distant metastasis are components of the AJCC stage group, the AJCC stage group was not included in the multivariate analysis. Similarly, as macroscopic growth pattern and microscopic tumor border were components of combined macroscopic growth pattern and microscopic tumor border status, macroscopic growth pattern and microscopic tumor border were excluded from the multivariate analysis. Based on multivariate analysis, Inf/Inf status (P = .02), higher tumor grade (P = .04), and the involvement of resection margin (P = .04) were identified as independent poor prognostic factors of PanNETs (Table 4).

Survival analysis of RFS was performed with 185 PanNET cases with clear resection margin, and 18 cases with involvement of resection margin were excluded (Table 5). Infiltrative macroscopic growth pattern (HR, 4.456; 95% CI, 1.821–10.91; P = .001), infiltrative microscopic tumor border (HR, 3.438; 95% CI, 1.839–6.427; P < .001), and Inf/Inf status (HR, 5.893; 95% CI, 2.755–12.61; P < .001) significantly predicted shortened RFS based on univariate analysis (Table 5). Other factors that predicted the RFS of patients with PanNETs included age (P < .001), symptoms (P = .04), tumor size (P < .001), tumor grade (G2, P = .02; G3, P < .001), pT classification (P < .001), lymph node metastasis (P < .001), and perineural invasion (P = .001).

AJCC stage group, lymphovascular invasion, and glucagon and gastrin expression were not included in the multivariate analysis, as these factors can affect pT classification, lymph node metastasis, and symptoms, respectively. Similarly, as macroscopic growth pattern and microscopic tumor border were components of combined macroscopic growth pattern and microscopic tumor border status, they were excluded from the multivariate analysis. As a result, Inf/Inf status (HR, 2.807; 95% CI, 1.097–7.184; P = .03) was the only independent prognostic factor significantly correlated with a short RFS.

Patients were further stratified based on tumor grade, pT classification, lymphovascular invasion, and nodal metastasis status because Inf/Inf status was the only independent poor prognostic factor based on multivariate analysis. Patients were well stratified by tumor grading based on the 2019 WHO classification scheme; patients with a higher grade had worse OS (5-year survival rate, G1, 88.5%; G2, 84.3%; G3, 25.0%; P < .001; Figure 5, A) and RFS (5-year survival rate, G1, 82.8%; G2, 65.3%; G3, 33.3%; P < .001; Figure 5, B). In G1, patients with PanNET having Inf/Inf status showed the tendency of worse OS (5-year survival rate, 85.9% versus 97.0%; P = .06; Figure 6, A) and significantly shorter RFS (5-year survival rate, 80.0% versus 91.2%; P = .04; Figure 6, B) than those with PanNET having other patterns or borders. Similarly, in higher grades (G2 and G3), patients with PanNET having Inf/Inf status showed significantly worse OS (5-year survival rate, 52.6% versus 94.4%; P < .001; Figure 6, C) and RFS (5-year survival rate, 35.0% versus 82.7%; P < .001; Figure 6, D) than those with PanNET having other growth patterns or borders.

In the lower pT (pT1–2) classification, patients with PanNET having Inf/Inf status had worse OS (5-year survival rate, 78.3% versus 96.2%; P = .001; Figure 7, A) and RFS (5-year survival rate, 71.9% versus 93.8%; P < .001; Figure 7, B) than those with PanNET having other patterns or borders. Also, in the higher pT (pT3–4) classification, patients with PanNET having Inf/Inf status showed significantly worse OS (5-year survival rate, 57.7% versus 96.0%; P = .02; Figure 7, C) and shorter RFS (5-year survival rate, 41.6% versus 71.4%; P = .02; Figure 7, D) than those with PanNET having other growth patterns or borders.

Subgroup analyses according to lymphovascular invasion status also revealed that patients with PanNET having Inf/Inf status had worse OS and RFS than those with other patterns or borders. First, in cases without lymphovascular invasion, patients with Inf/Inf showed significantly worse OS than those with other patterns or borders (5-year survival rate, 80.4% versus 96.5%; P = .01; Supplemental Figure 1, A [see supplemental digital content containing 8 figures at https://meridian.allenpress.com/aplm in the January 2023 table of contents]). Similarly, patients with Inf/Inf status showed significantly shorter RFS than those with other patterns or borders (5-year survival rate, 67.7% versus 89.8%; P < .001; Supplemental Figure 1, B). In cases with lymphovascular invasion, patients with Inf/Inf status showed significantly worse OS than those with other patterns or borders (5-year survival rate, 57.8% versus 94.4%; P = .04; Supplemental Figure 1, C). Similarly, patients with Inf/Inf status showed significantly shorter RFS than those with other patterns or borders (5-year survival rate, 47.0% versus 80.8%; P = .04 Supplemental Figure 1, D).

In the pN0 classification, patients with PanNET having Inf/Inf status had worse OS (5-year survival rate) than those with other patterns or borders (73.7% versus 96.0%; P < .001; Supplemental Figure 2, A). Similarly, patients with Inf/Inf showed significantly shorter RFS than those with other patterns or borders (5-year survival rate, 67.5% versus 90.0%; P < .001; Supplemental Figure 2, B). In the pN1–2 classification, there was no statistically significant difference in OS (5-year survival rate, 56.4% versus 100.0%; P = .29, Supplemental Figure 2, C) or RFS (37.5% versus 53.3%; P = .43; Supplemental Figure 2, D) between Inf/Inf status and other patterns or borders. These results might be due to the small number of these cases.

To the best of our knowledge, this study is the first to assess the clinicopathologic significance and prognostic value of both macroscopic growth pattern and microscopic tumor border using a large number of PanNET cases. Infiltrative macroscopic growth pattern and infiltrative microscopic tumor border were both significantly correlated with aggressive clinicopathologic factors, such as tumor being nonfunctioning, larger size, higher grade, higher pT classification, higher AJCC stage, presence of nodal metastasis, lymphovascular and perineural invasions, and no insulin expression. Patients with infiltrative macroscopic growth pattern and infiltrative microscopic tumor border showed significantly worse OS and shorter RFS than those with expansile macroscopic growth pattern and pushing microscopic tumor border, respectively. Furthermore, when these 2 factors were combined, infiltrative combined macroscopic growth pattern and microscopic tumor border status was identified as an independent prognostic factor for poor OS and the only independent predictor for short RFS. In addition, interobserver agreement could be increased substantially by combining both factors. Finally, combined macroscopic growth pattern and microscopic tumor border status could predict survival and tumor recurrence in patients based on tumor grading and pT classification. Our findings suggest that assessing both macroscopic growth pattern and microscopic tumor border in daily pathology practice is a simple, feasible, and economic strategy for predicting different biological features and the prognosis of patients with PanNETs.

The growth pattern based on gross examination and tumor border based on microscopic examination could provide important information on the biological behavior of mass-forming lesions and have previously been emphasized as valuable prognostic factors in cancers of several other organs.^{24,27,40,45–50}  In the case of macroscopic growth pattern, the infiltrative pattern of cervical cancer on preoperative magnetic resonance images was a significant independent factor for poor OS.⁵⁰  Similarly, hepatocellular carcinoma with an infiltrative growth pattern was associated with shorter RFS with more frequent vascular invasion than that with other gross patterns.^46,47  Zlobec et al²⁴  demonstrated the additive effect of microscopic tumor border for stratifying OS in patients with stage II colorectal cancer, and adverse molecular alteration, such as BRAF^V600 mutation, was correlated with infiltrative tumor border of colorectal cancer.⁴⁰  Similarly, infiltrating tumor border of breast cancer and papillary thyroid carcinoma was correlated with aggressive behavior with lymph node metastasis.^48,49  In particular, a 3-tiered, pattern-based classification system of tumor infiltration in endocervical adenocarcinoma, named the Silva system, is widely used today in daily diagnosis of endocervical adenocarcinoma to better predict patients' prognosis and aid physicians in deciding further treatment, especially lymphadenectomy.²⁷ 

Likewise, we attempted to stratify patients with PanNET by the tumor border based on pattern of invasion to correctly stratify patients who need additional intervention after surgical resection. There have been few studies on the immune microenvironment of PanNETs at the invasive margin,^45,51  and infiltrative microscopic tumor border of PanNETs was emphasized previously in a few studies,^36–38  with correlation of adverse clinicopathologic factors, such as higher risk of metastasis, recurrence, and lymphovascular invasion. However, there have been no previous reports systematically reviewing both the macroscopic and microscopic tumor border of PanNETs for survival analysis and correlation with various clinicopathologic factors. We sampled all areas of the tumor with different appearance, and at least one representative section per 1 cm of the tumor was submitted at the time of gross examination. Our results demonstrated that 1 section per centimeter was enough for microscopic evaluation of the tumor border. However, there may not be a single answer, because the appropriate number of sections depends on the type of specimen. Further study is required for solving this issue.

In the present study, PanNET with an infiltrative macroscopic growth pattern was correlated with tumors with the following features: nonfunctioning, larger size, higher tumor grade, higher pT classification, lymph node metastasis, distant metastasis, higher AJCC stage grouping, lymphovascular and perineural invasions, and no insulin expression. Similarly, PanNET with an infiltrative microscopic tumor border was more frequently observed in patients of older age and in men, and infiltrative border status was associated with tumors with the following features: nonfunctioning, larger size, higher tumor grade, higher pT classification, lymph node metastasis, higher AJCC stage grouping, lymphovascular and perineural invasions, involvement of resection margin, and no insulin expression. These clinicopathologic factors, including patients with older age, tumors being nonfunctioning, higher tumor grade, lymph node metastasis, and lymphovascular invasion, were reported to be independent poor prognostic factors in patients with PanNETs in previous studies.^12,13,16  To our knowledge, our study is the first to assess the combined macroscopic growth pattern and microscopic tumor border in PanNETs. Our observations of shorter survival in patients with microscopic infiltrative border were consistent with those of the previous studies of PanNETs.^36–38  In addition, we have confirmed that interobserver agreement in assessing microscopic tumor border can be substantially increased by combining macroscopic growth pattern with microscopic tumor border. By combining macroscopic growth pattern and microscopic tumor border, additional information can be provided; both features when combined can be used as an independent prognostic factor of the survival and tumor recurrence in patients with PanNET.

Our study has several limitations. First, only a few cases in each of the prognostic groupings, especially patients with G3, pT4, pN1 to 2, and AJCC stage IV tumors, were included. Nonetheless, based on subgroup analysis of our cohort, relatively accordant survival and recurrence data were found. Further, the importance of the tumor border was highlighted in further stratification of these groups. Second, this was a retrospective study with a limited number of available gross images of the cross section and tumor sections of the PanNETs, which may have led to missing a few cases with focal infiltrative growth pattern. Last, the macroscopic growth pattern and microscopic tumor border cannot be assessed in endoscopic ultrasound-guided fine needle aspiration biopsy specimens. Therefore, further studies on the role of preoperative imaging and combined macroscopic growth pattern and microscopic tumor border status in patients with PanNETs are needed.

In summary, PanNETs with combined infiltrative macroscopic growth pattern and infiltrative microscopic tumor border were observed in approximately one-quarter of the PanNET cases evaluated in this study and were associated with aggressive behavior and short OS and RFS. Therefore, assessment of the macroscopic growth pattern and microscopic tumor border status, which is a simple, reproducible, and cost-saving method, can provide additional information and improve current prognostic classification of the recurrence and survival of patients with surgically resected PanNETs.