Noninvasive markers to predict malignant intraductal papillary mucinous neoplasms of the pancreas.

204 Background: Despite guidelines to preoperatively distinguish malignant from benign intraductal papillary mucinous neoplasms (IPMN), many patients undergo invasive testing and morbid pancreatic resection but ultimately have benign disease on pathology. A blood-based microRNA signature (SIG, including miR-200a-3p, miR-1185-5p, miR-33a-5p, miR-574-4p, and miR-664b) has been shown to preoperatively discriminate malignant from benign IPMN, with expression lower in malignant IPMN. This study was designed to develop a model to improve preoperative prediction of IPMN pathologic status combining radiographic markers and SIG values. Methods: An institutional database was used to identify patients undergoing resection for IPMN (2006-2011) with preoperative computed tomography (CT) scans and SIG values. CTs were read by a single radiologist blinded to pathology to assess predetermined radiographic features. The outcome was malignant pathology (MP, invasive carcinoma and high-grade dysplasia) vs. benign pathology (BP, low- and moderate-grade). Results: Of 38 eligible patients, 20 (53%) had MP and 18 (47%) had BP. 72% with MP had main pancreatic duct (PD) involvement vs. 20% with BP, p = 0.003. Median cyst size was higher in the MP group (3.9 vs. 2.8cm), p = 0.018. 83% of those with MP had ≥ 1 “high-risk stigmata” (PD size ≥ 10mm, enhancing solid component, or obstructive jaundice), vs. 15% of those with BP (p < 0.001), yet ≥ 1 “worrisome” feature (acute pancreatitis, PD size 5-9mm, cyst size > 3cm, thickened enhanced cyst walls, or non-enhanced mural nodules) was not associated with malignancy (p = 0.734). SIG was significantly lower in the MP group, p < 0.001. Multivariate logistic regression analyses revealed that high risk stigmata and SIG retained significance (43.0 [4.64-398.8], p = 0.001 and 0.30 [0.10-0.86], p = 0.026, respectively). The area under the receiver operating characteristic curve resulted in 0.950 for the model with both variables, compared to 0.841 and 0.836 for each variable independently. Conclusions: Combining high-risk stigmata from preoperative CT scans with a blood-based miRNA genomic classifier may improve the ability to noninvasively predict IPMN status preoperatively.