Abstract / Summary
Purpose: Pancreatic ductal adenocarcinoma (PDAC) has high rates of early recurrence (ER), occult metastasis, and poor survival despite curative-intent therapy. Predictive factors and models span clinical, biomarker, imaging, radiomics, pathologic, and molecular domains, but clinical readiness remains uncertain.
Methods: We performed a systematic review in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses. Five databases were searched for studies evaluating predictors or models of PDAC recurrence, metastasis, lymph node metastasis (LNM), occult metastatic disease, disease-free survival (DFS) or recurrence-free survival, or recurrence patterns. Studies were screened and coded for end point, modality, model type, validation, discrimination, calibration, clinical utility, and predictor timing.
Results: Of 234 full-text records, 208 studies were included, comprising 144 prediction models and 64 predictor/prognostic factor studies. Computed tomography/radiomics (141/208), clinical/electronic health record variables (104/208), and serum/urine biomarkers (46/208) were the most common modalities. External validation, calibration, and decision-curve analyses were reported in 52, 36, and 38 model studies, respectively. AUCs ranged from 0.70 to 0.92 for ER, 0.75-0.95 for LNM, and 0.71-0.88 for liver or occult metastasis; DFS/relapse-free survival models had C-indices of 0.62-0.82. One externally validated occult-metastasis model had a C-statistic of 0.56. Model maturity was limited by retrospective design, heterogeneous end points, incomplete calibration reporting, and postoperative variables in preoperative models.
Conclusion: Prediction models for PDAC recurrence and metastatic progression are rapidly expanding, particularly through radiomics, machine learning, and multimodal modeling. Clinical translation requires decision-aligned design, transparent reporting, external validation, calibration, interpretability, and prospective evaluation of clinical utility.