All patient cases were included in statistical analysis comparing concordance rates of (1) Ki67 grading between EUS-FNA cytology and surgical pathology specimens and (2) Ki67 grading and mitotic rate grading of surgical pathology specimens.

All patients, except one patient with synchronous metastasis during diagnosis, were included in the Kaplan–Meier estimator for disease recurrence. Fisher’s exact test was used for univariate analysis to determine the risk factors associated with disease-related death and disease recurrence. One patient had a NF-PNET resection with disease recurrence at seven years after surgical resection. This patient’s second tumor data was excluded from receiver operating characteristic (ROC) curve analysis and Fisher’s exact test. P-values less than 0.05 were regarded as statistically significant. Given our sample size, odds ratios were calculated with a two-by-two table instead of logistic regression.

All data were analyzed with SPSS 24.0 for Windows.

From 2005 to 2015, 36 patients (62% female; median age, 62 years; range, 38–83 years) with NF-PNETs underwent both EUS-FNA and surgical resection of their tumors. One patient was included twice as she developed recurrence seven years after her initial surgical resection; she underwent both EUS-FNA and surgical resection during recurrence.

Tumor size ranged from 7 to 170 mm (median, 25 mm; mean, 40 mm). Thirteen tumors (35.1%) were located in the head, 23 (62.2%) were located in the body and tail, and 1 was found throughout the entire pancreas (size, 65 mm). At the time of resection, positive lymph nodes were found in 11 (29.7%) cases, lymphovascular invasion was present in 17 (45.9%), and synchronous metastasis was present in 1 (2.7%) case (Table 1).

Needle sizes used for EUS-FNA sampling included 19 G, 22 G, and 25 G. In 16 cases, only 25 G needles were used; in 8 cases, only 22 G needles were used; in the remaining cases, a combination of needle sizes was used. Number of passes ranged from 1 to 9 passes, with a mean and median of 5 passes. Time from the EUS-FNA to surgical resection ranged from 3 days to 115 days, with an average of 31.2 days.

Five patients were lost to follow-up within three months following resection. For those who remained in follow-up, average follow-up was 49.9 months (range, 4.6 to 134.1 months) and average disease-free survival was 53.9 months.

One patient had high-grade Ki67 and synchronous metastasis during diagnosis and died eventually. Seven patients developed disease recurrence following resection (range, 3 to 84 months). Six of these patients had lymphovascular invasion and four patients concomitantly had positive lymph nodes during initial resection. In two cases with high-grade Ki67, time to disease recurrence was ≤6 months and disease-related death was the eventual clinical outcome. This is compared to four low-grade cases and one intermediate-grade Ki67 case that had disease recurrence between 16 and 84 months (Table 2).

Twenty-seven out of 37 cases (73.0%) had concordant Ki67 grades between cytology and histology (Table 3). Concordance was the highest in (20/24) low-grade tumors and (4/5) high-grade tumors, with positive predictive values of 80.0% and 100% respectively. Only 3 of 8 cases of intermediate tumors were concordant, with a positive predictive value of 37.5%.

In the ten discordant cases, four tumors graded as intermediate by EUS-FNA were downgraded to low after surgical resection; five tumors graded as low by EUS-FNA were upgraded to intermediate after surgical resection; and one tumor that was graded as intermediate by EUS-FNA was upgraded to high after surgical resection.

In all cases that were upgraded, tumor size was equal to or greater than 25 mm. In cases that were downgraded, tumor size ranged from 10 mm to 140 mm. Number of passes and location did not differ significantly in discordant cases. In cases that were later upgraded, number of passes ranged from three to nine. In cases that were later downgraded, number of passes ranged from two to eight (Table 4). Time between EUS-FNA and surgical resection did not differ significantly between tumors with concordant versus discordant Ki67 grading.

Thirty out of 37 cases (81.1%) had concordance between Ki67 grading and mitotic rate grading in surgical resected specimens. Concordance was the highest in (22/24) low-grade tumors and (5/5) high-grade tumors, with positive predictive values of 81.5% and 83.3% respectively. Only 3 out of 8 intermediate-grade tumors were concordant (Table 3).

Three patients had disease-related death during the follow-up period. All three patients had high-grade Ki67 on both EUS-FNA cytology and surgical resection specimen histology. One had synchronous metastasis during diagnosis with time to disease-related death at 46 months; the other two had disease recurrence within 6 months of surgical resection and had disease-related death at 8 months and 18 months, respectively.

Because time to disease-related death can be long, we defined “poor prognosis” as either disease-related death or disease recurrence and investigated factors leading to those outcomes. ROC curve analysis showed that tumor size ≤16.5 mm would be able to exclude all NF-PNETs with poor prognosis, while tumor size ≤22.5 mm would be able to exclude 87.5% of NF-PNETs with poor prognosis, area under the curve=0.743. Univariate analysis confirmed EUS-FNA Ki67 ≥20% (p=0.028), surgical Ki67 ≥20% (p=0.028), and mitotic rate >20 per 10 HPF (p=0.005) to be significantly associated with poor prognosis (Table 5).

Because only high-grade Ki67 NF-PNETs were significantly predictive of poor prognosis, our Kaplan Meier analysis investigated time to disease recurrence for high-grade versus low-and intermediate-grade tumors based on surgical Ki67 grading. The Kaplan Meier estimator demonstrated a significantly shorter amount of time to disease recurrence for high-grade tumors compared to low- and intermediate-grade tumors (Fig. 1).

Those with lymphovascular invasion during surgical resection were more likely to have poor prognosis (37.5%) compared to those without lymphovascular invasion (10%). However, this association did not meet statistical significance (p=0.103). Similarly, tumor size >22.5 mm tended to correlate with poor prognosis (33.3%) compared to those with smaller tumor size ≤22.5 mm (6.7%), but this association was not statistically significant (p=0.104). Positive lymph nodes were not associated with poor prognosis.

This retrospective study is the largest single-center series (n=37) investigating Ki67 concordance between EUS-FNA cytology and surgical pathology specimens in NF-PNETs. Prior series have shown Ki67 concordance rates ranging between 69.2% and 92% in all PNETs [8,28-34]. However, only Larghi et al. have validated EUS-FNA as an accurate technique for Ki67 determination in NF-PNETs specifically [29]. Our study reports a 73% concordance of Ki67 between cytological and resected tissue samples. Concordance rates were the highest in low- and high-grade PNETs based on surgical Ki67. However, similar to previous studies, intermediate-grade tumors had a lower concordance rate between Ki67 grading on EUS-FNA cytology and surgical resection histology [8,25,27]. The lower Ki67 concordance rate in our intermediate-grade NF-PNETs could be due to several causes including the following: (1) intratumoral heterogeneity that is reported to be higher in intermediate-grade surgical Ki67 tumors than in low-grade surgical Ki67 tumors [25]; (2) increased tumor size, which may contribute to increased intratumoral heterogeneity [27]; or (3) lower cell count from EUS-FNA sampling, as Hasegawa et al. noted increased Ki67 concordance rates from 77.8% to 90% when patients with <2,000 tumor cells were excluded [8]. Overall, our Ki67 concordance rate was consistent with prior studies considering both functioning and NF-PNETs, but on the lower end of the range of previously reported concordance rates. This finding may be attributed to increased intratumoral heterogeneity related to larger tumor size, as our average tumor size was 40 mm. Another possibility is that NF-PNETs may have higher intratumoral heterogeneity compared to functioning PNETs, thus contributing to lower Ki67 concordance rates.

Due to increased use of cross-sectional imaging, incidental detection of NF-PNETs has increased. Risk stratification of these asymptomatic but clinically unpredictable tumors is important in determining management. Many studies have reported elevated Ki67 as a poor prognostic marker for PNETs. Hamilton et al. found that tumors with surgical Ki67 >9% were more likely to have disease recurrence and decreased overall survival, while Bettini et al. reported surgical Ki67 >5% to be associated with disease-related mortality [20,22]. Our study confirms elevated Ki67 as a poor prognostic factor for disease recurrence and disease-related death in both EUS-FNA and surgical samples. Despite lower concordance rates for intermediate-grade NF-PNETs, overall concordance was strong enough that EUS-FNA and surgical Ki67 ≥20% were both associated with disease-related death, with time to disease-related death ranging from 8 months to 46 months despite surgical resection. Additionally, because intermediate-grade NF-PNETs demonstrated lower Ki67 concordance and are at risk of being under-graded, NF-PNETs with intermediate-grade Ki67 on EUS-FNA should strongly be considered for surgical resection.

Our data shows that tumor size ≤16.5 mm and ≤22.5 mm would be able to exclude 100% and 87.5% of NF-PNETs with poor prognosis, respectively. While increased tumor size had a greater odds ratio of poor prognosis, this association was not statistically significant; thus, we do not recommend using tumor size >16.5 mm to define tumors that warrant surgical resection. Based on our data, all NF-PNETs whose EUS-FNA Ki67 grading was upgraded after surgical resection were ≥25 mm. This parallels the Fujimori et al. study where tumors ≥20 mm had lower rates of concordance between Ki67 obtained from EUS-FNA versus surgical pathology specimen [27]. These findings strongly support the growing clinical practice and the ENETS’s guidelines of observing NF-PNETs with low-grade Ki67 on EUS-FNA up to 20 mm [18]. Furthermore, our data suggest that a higher threshold of 25 mm for surgery may be acceptable for low-grade EUS-FNA tumors, as using this cutoff would have excluded any NF-PNETs that should have been resected.

Time to disease recurrence was notably shorter for patients with high-grade surgical Ki67 (3 and 6 months) than for low-and intermediate-grade surgical Ki67 (all >12 months). Of note, approximately 50% of our patients had undergone surgical resection within the last five years; therefore, our results may not capture all cases that will develop disease recurrence and disease-related death. We suggest that patients with NF-PNETs with surgical Ki67 ≥20% should be monitored at closer intervals post-operatively, while those with low- and intermediate-grade surgical Ki67 may need more prolonged post-operative monitoring for disease recurrence.

In addition to Ki67, our study demonstrated mitotic rate as a separate risk factor for disease recurrence and disease-related death. Mitotic rate is a separate measure used by the ENETS and WHO to grade PNETs. Thus far, studies investigating concordance between Ki67 index and mitotic rate grading have produced conflicting results [8,28-34]. We demonstrate 81.1% concordance rate between Ki67 and mitotic rate grading in surgical resection specimens. Additionally, mitotic rate grading paralleled Ki67 grading in predicting poor prognosis, as tumors with mitotic rate >20 per 10 HPF were associated with both disease-related death and disease recurrence. Given the high concordance between surgical Ki67 and mitotic rate as well as the independent risk factor of high-grade mitotic rate for disease recurrence and disease-related death, we support using mitotic rate adjunctively with surgical Ki67 for post-operative surveillance.

Lastly, we found that tumors with lymphovascular invasion during surgical resection had a poor prognosis. This trend was not statistically significant, likely due to small sample size and short duration of follow-up. However, given this finding, patients with NF-PNETs with lymphovascular invasion during surgery may benefit from both shorter interval and prolonged course of post-operative surveillance.

The strengths of this study include its size, uniformity, and correlation with natural history. As the nature of the study was retrospective, and all the data were compiled at one medical center, we had access to relevant imaging, EUS-FNA findings, surgical pathology results, as well as a majority of follow-up data. The drawbacks of this study include the relatively short follow-up for disease recurrence and disease-related death, due largely in part to the prolonged disease process. Given that previous studies used different Ki67 and tumor size cutoffs to predict poor prognosis, future studies should investigate whether different Ki67 and tumor size cutoffs would better risk stratify NF-PNETs. Studies with longer follow-up may elucidate the association of low- and intermediate-grade tumors with time to disease recurrence and disease-related death. Finally, investigation of factors leading to discordance of Ki67 between EUS-FNA and surgical pathology (i.e., larger tumor size and increased heterogeneity) can improve our understanding of obtaining optimal tumor samples.

In summary, our study demonstrates that Ki67 indices obtained by EUS-FNA can be used reliably to predict surgical Ki67 indices in NF-PNETs. Our results show that (1) NF-PNETs with high-grade Ki67 on EUS-FNA had poor prognosis despite surgical resection; (2) NF-PNETs with intermediate-grade Ki67 on EUS-FNA should be strongly considered for surgical resection; (3) NF-PNETs with low-grade Ki67 on EUS-FNA can be safely observed up to tumor size 20 mm. Post-operatively, patients with low- and intermediate-grade surgical Ki67 tumors may benefit from prolonged surveillance. Mitotic rate and presence of lymphovascular invasion during resection should be considered in post-operative surveillance.