Background: Data on the reliability of the Ki-67 index and grading calculations from endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) of pancreatic neuroendocrine tumors (PanNETs) are controversial. We aimed to assess the accuracy of these data compared with histology.

Methods: Cytological analysis from EUS-FNA in patients with suspected PanNETs (n = 110) were compared with resection samples at a single institution. A minimum of 2000 cells were considered to be adequate for grading. Correlation and agreement between cytology and histology in grading and Ki-67 values, respectively, were investigated. Secondary outcomes included the diagnostic performance of EUS-FNA.

Results: EUS-FNA samples were adequate for PanNET diagnosis and PanNET grading in 98/110 (89.1 %) and 77/110 (70.0 %) patients, respectively; thus, 77 samples were adequate for comparing cytology vs. histology. There were 67 (62.0 %), 40 (36.4 %), and 1 (0.9 %) patients with a final diagnosis of G1, G2, and G3 tumors, respectively. EUS-FNA grading was concordant with surgical pathology in 81.8 % of patients; under- and overgrading occurred in 15.6 % and 2.6 %, respectively. The overall level of agreement for grading was moderate (Cohen's κ = 0.59, 95 % confidence interval [CI] 0.34 - 0.78). Spearman's rho for Ki-67 in tumors ≤ 20 mm and > 20 mm was strong and moderate, respectively (rho = 0.68, 95 %CI 0.47 - 0.83; rho = 0.59, 95 %CI 0.35 - 0.75). The Bland - Altman plot showed that the Ki-67 values were comparable and reproducible between the two measurements.

Conclusions: Although they were not available for a significant number of patients, grading and Ki-67 values from cytology correlated with histology moderately to strongly.