Abstract
PURPOSE: Patients with unresectable pancreatic cancer (URPC) have poor prognoses and heterogeneous responses to systemic chemotherapy. Existing staging systems show limited accuracy for prognostic assessment. We aimed to develop and validate a radiopathomics signature for pancreatic cancer (RPSPC) to estimate overall survival (OS) and evaluate chemotherapy benefit. METHODS: Ninety-eight patients with URPC were enrolled retrospectively and divided into training (n = 69) and validation (n = 29) cohorts. Radiomics features were extracted from contrast-enhanced computed tomography, and pathomics features were obtained from biopsy-derived whole-slide images. RPSPC was developed to predict OS, and its association with OS was assessed. Hyperparameters were optimized by five-fold cross-validation in the training cohort. The concordance index (C-index) and the AUC were calculated in both cohorts. Patients were stratified into high- and low-RPSPC groups to assess chemotherapy benefit. RESULTS: RPSPC was independently associated with OS in the training cohort (hazard ratio [HR], 2.636; P = .003). The nomogram incorporating RPSPC and carbohydrate antigen 19-9 level achieved C-indices of 0.793 in the training cohort and 0.792 in the validation cohort. For 1-year survival prediction, the nomogram exhibited AUCs of 0.906 and 0.859 in the training and validation cohorts, respectively. In the total cohort, patients with high RPSPC had significant survival benefit from systemic chemotherapy (HR, 0.492; P = .020), whereas patients with low RPSPC did not have significant survival benefit (HR, 0.621; P = .176). CONCLUSION: RPSPC could serve as an independent prognostic factor for patients with URPC and might help identify those who benefit from chemotherapy.