Introduction Distinguishing benign from potentially malignant pancreatic cystic lesions has important prognostic and therapeutic implications. Diagnosis is achieved through a combination of history, imaging, biochemical, endoultrasonographic (EUS) and cyst aspirate (CA) analysis. We evaluated the accuracy of suspicious and high risk variables identified at EUS and CA analysis to predict premalignant or malignant histopathology confirmed at surgery.

Method Patients who underwent EUS prior to surgery were selected. Data on age, sex, cyst location, suspicious (S) or high risk (HR) EUS or CA features, diagnosis (grouped as benign vs premalignant/malignant) at EUS (EUS-D), post CA analysis (CA-D) and combined EUS and CA analysis (EUS&CA-D) were collected. These were compared with final histological diagnosis (H-D) from surgery. High-risk (HR) variables were defined as C4/C5 diagnosis or carcinoma embryonic antigen (CEA) >192 ng/ml on CA, and cyst size >30 mm, PD dilatation >10 mm, mural nodules or mixed solid/cystic components at EUS Suspicious (S) variables included HR features or mucin, C3 diagnosis or amylase >1000 U/L on CA, and pancreatic duct (PD) dilatation or cyst wall thickening on EUS. Variables were assigned a score of 1 when present, and combined with either EUS-D (EUS-D-S or EUS-D-HR) or CA-D (CA-DS or CS-D-HR). A single positive variable defined positivity in groups. EUS and CA outcomes were then evaluated to identify the sensitivity, specificity, area under the ROC curve (AUC) and likelihood ratio (LR) to predict H-D.

Results 38 patients were identified (mean age 60.4 years, range 19–81; 52.6% female). Histology identified benign (18.5%; retention cysts, accessory spleen, chronic pancreatitis, hydatid, serous cystic neoplasm), premalignant (60.5%; intraductal papillary mucinous neoplasm (IPMN), mucinous cystadenoma, solid pseudopapillary tumour) or malignant disease (21.1%; adenocarcinoma, adenosquamous carcinoma, neuroendocrine tumour). Analysis (AUC and LR) ranked by AUC confirmed EUS-D-S significantly predicted H-D (0.81, p = 0.022; 5.87) with a sensitivity and specificity of 83.7% and 85.7% respectively. This was followed by EUS-D-HR (0.68; 4.29; p = 0.187), CS-D-HR (0.64; 1.83; p = 0.306), EUS&CA-D-HR (0.62; 1.81; p = 0.374) and CS-D-S (0.52; 1.04; p = 0.89). EUS&CA-D-S did not predict H-D (0.5; 1.16; p = 1.0).

Conclusion EUS can accurately predict final H-D in patients cystic pancreatic lesions with excellent sensitivity and specificity when modelled appropriately. Interestingly, the accuracy of this method diminished when combined with CA variables.

Disclosure of interest None Declared.