Machine learning-based construction of a programmed cell death-related model reveals prognosis and immune infiltration in pancreatic adenocarcinoma patients.

Pancreatic adenocarcinoma (PAAD) is a highly lethal malignancy with limited effective prognostic biomarkers. In this study, 1,034 samples from TCGA-PAAD, GSE62452, GSE28735, GSE183795, and ICGC cohorts were systematically integrated to identify key programmed cell death-related genes (PCDRGs) associated with patient prognosis. Differential expression analysis and Univariate Cox regression analysis identified 17 candidate PCD-related genes significantly associated with overall survival. Using a comprehensive machine learning framework involving 117 algorithmic combinations under a Leave-one-out cross-validation (LOOCV) strategy, we identified the StepCox[both] + Ridge as the best algorithms composition to construct a prognostic model based on six PCDRGs, ITGA3, CDCP1, IL1RAP, CLU, PBK, and PLAU. The model was validated to have robust predictive performance. Risk scores were significantly correlated with clinical features, immune microenvironment characteristics, and chemotherapeutic sensitivity. High-risk patients exhibited worse prognosis and immunosuppressive infiltration patterns. Furthermore, consensus clustering identified two PAAD molecular subtypes with distinct PCDRGs expression patterns and survival outcomes. A nomogram integrating risk score and clinical variables exhibited strong prognostic accuracy for 1-, 3-, and 5-year survival prediction. In summary, we established and validated a PCD-related prognostic signature that effectively stratifies PAAD patients by clinical outcome, immune contexture, and therapeutic response, providing novel insights for personalized management strategies.