ABSTRACT Background Pancreatic cancer remains a highly lethal malignancy due to late diagnosis and the lack of effective non‐invasive biomarkers for detection, further complicated by biological heterogeneity, including intraductal papillary mucinous neoplasm (IPMN) and IPMN‐associated invasive carcinoma (IPMC). Methods Serum samples from a discovery cohort ( n = 60), including patients with IPMN, IPMC, and pancreatic ductal adenocarcinoma (PDAC), were analyzed using LC–MS/MS‐based serum proteomic profiling to identify candidate biomarkers. Feature selection was performed using ANOVA, receiver operating characteristic analysis, and permutation feature importance. The selected markers were further evaluated in an independent validation cohort ( n = 40). Results In the discovery cohort, the model demonstrated robust performance, with leave‐one‐out cross‐validation AUROC values ranging from 0.814 to 1.000 across classifications of healthy controls and disease groups (IPMN, IPMC, and PDAC). In the independent validation cohort, the six‐marker panel (CALR, FCN1, MBL2, SPP1, TAGLN2, and VWF) achieved an AUROC of 0.962 with a specificity of 95.0% for distinguishing healthy controls from IPMC, and an AUROC of 0.856 with a specificity of 90.0% for healthy controls vs. PDAC. Conclusions These findings demonstrate the robustness and clinical potential of the six‐protein panel as a non‐invasive diagnostic tool for pancreatic cancer.
Jung et al. (Fri,) studied this question.