Grading pancreatic neuroendocrine neoplasms by Ki-67 staining on cytology cell blocks: manual count and digital image analysis of 58 cases

Introduction Controversy remains as to whether Ki-67 labeling on endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) cell blocks (CBs) can be used to grade pancreatic neuroendocrine neoplasms (PanNENs) reliably and what would be the best methodology for doing so. Materials and methods A retrospective search identified cases with both EUS-FNA and correlating surgical pathology (SP) available. 58 CBs (>100 tumor cells) were identified. Ki-67 labeling on CB was counted using both manual count (MC) and digital image analysis (DIA, ImmunoRatio software). The grades determined on CB were compared with those on corresponding SP reports (kappa statistics). The correct grading rates between MC and DIA were compared (McNemar's test). Total tumor cell number (TTCN, cutoffs: 500 and 2000) was used as a stratification factor (chi-square tests for trend). Results Of 58 cases, the SP grade distribution was 31 grade 1, 23 grade 2, and 4 grade 3. Overall, the grading concordance of CB MC and SP was higher than that of CB DIA and SP (McNemar's P-value = 0.022). Compared with SP, CB MC correctly graded 69% (40 of 58) [κ = 0.44 (95% CI: 0.21, 0.66)]; CB DIA correctly graded 55% (32 of 58) [κ = 0.19 (95% CI: 0.04, 0.42)]. Grade 1 tumors had the highest concordance. Although the correct grading rate improved as TTCN increased, no statistical significance was found. Conclusion Although Ki-67 labeling on CB can be used to grade PanNENs, limitation exists, particularly for grade 2 tumors. The method of counting makes a difference: MC is more accurate than DIA. The major reasons of discordance include non-tumor cell contamination and insufficient sampling.