Objective: This study aimed to develop and validate a preoperative predictive model to identify patients at high risk of early recurrence (ER), with a view to establish a framework for biological borderline resectability of non-functioning pancreatic neuroendocrine tumors (NF-PanNETs).
Background: Radical surgery is curative for most localized NF-PanNETs, but a subset of patients experiences ER. No standardized criteria define preoperative high-risk disease.
Methods: A retrospective multicentric study was conducted at 3 tertiary centers. Patients undergoing curative resection for localized NF-PanNETs were included, and preoperative clinicopathologic and imaging variables were analyzed. ER was defined as a recurrence within 24 months. A classification tree model was developed, and performance was assessed using the area under the curve (AUC) of the receiver operating characteristic curve.
Results: A total of 496 patients were analyzed, with 290 in the derivation cohort and 206 in the validation cohort. ER occurred in 55 patients (11%), including 26 (9%) in the derivation and 29 (14%) in the validation cohort. The median disease-free survival for ER patients was 16 months (interquartile range: 10-20 months). Neoplastic venous thrombosis was the strongest predictor of ER, with an ER probability of 71%. Among patients without venous thrombosis, those with a Ki-67 index ≥5% and tumor size ≥3 cm had an ER probability of 41% in case of adenopathy and 19% otherwise. The model achieved an AUC of 0.91 in the derivation cohort and 0.84 in the validation cohort.
Conclusions: This externally validated model provides a reliable preoperative tool to identify NF-PanNETs at high risk of ER and introduces the concept of biological borderline resectable NF-PanNETs.
Keywords: Ki-67 index; biological borderline resectable; disease-free survival; early recurrence; non-functioning pancreatic neuroendocrine tumors; preoperative predictive model; venous thrombosis.
Copyright © 2025 Wolters Kluwer Health, Inc. All rights reserved.