Cytologic predictors of malignancy in bile duct brushings: a multi-reviewer analysis of 60 cases.

Diagnosing malignancy in bile duct brushings is highly challenging. Seven reviewers of variable backgrounds and levels of participation in bile duct brushing sign out blindly reviewed 60 specimens (30 malignant with histologic confirmation and 30 benign (15 stented) with resection or ≥18 months of uneventful follow-up), testing the utility of 14 malignant characteristics. Eleven characteristics were statistically significantly associated with malignancy including 3-dimensional clusters (63% in malignant vs 3% in benign, odds ratio 50, P=0.0003), pleomorphism (62 vs 3, odds ratio 48, P=0.0004), 2-cell population (60% vs 3, odds ratio 44, P=0.0005), chromatin pattern (hypo/hyperchromasia) changes (70% vs 7%, odds ratio 33, P<0.0001), high nuclear-to-cytoplasmic ratio (48 vs 3%, odds ratio 27, P=0.0023), cytoplasmic vacuoles (43 vs 3%, odds ratio 22, P=0.0042), nuclear irregularity (70 vs 10%, odds ratio 21, P<0.0001), cellular discohesion (38 vs 3%, odds ratio 18, P=0.0082), hypercellularity (23% vs 0), nuclear molding (20% vs 0) and prominent nucleoli (21% vs 0). Necrosis and infiltrating inflammation were not helpful in identifying malignancy ('neutrophil cannibalism' was noted in 43% malignant); 21/30 (70%) malignant brushings had ≥3 malignant characteristics, while 23 (77%) benign brushings had none. Of 20 brushings with ≥4 characteristics, 1(5%) proved benign and showed detachment atypia, a close malignant mimicker in brushings. Identification of 3 characteristics maximized the combined sensitivity (70%), specificity (97%) and accuracy (83%), but sensitivity dropped as number of characteristics increased. Identification of 3/11 characteristics (3-dimensional clusters, pleomorphism, high nuclear-to-cytoplasmic ratio, nuclear irregularity, hypercellularity, discohesion, chromatin changes, vacuoles, prominent nucleoli, molding and 2-cell population) improves pathologists' overall performance greatly.