We systematically developed a prognostic model for pancreatic cancer that was compatible across different transcriptomic platforms and patient cohorts. After performing quality control measures, we used seven microarray datasets and two RNA sequencing datasets to identify consistently dysregulated genes in pancreatic cancer patients. Weighted gene co-expression network analysis was performed to explore the associations between gene expression patterns and clinical features. The least absolute shrinkage and selection operator (LASSO) and Cox regression were used to construct a prognostic model. We tested the predictive power of the model by determining the area under the curve of the risk score for time-dependent survival. Most of the differentially expressed genes in pancreatic cancer were enriched in functions pertaining to the tumor immune microenvironment. The transcriptome profiles were found to be associated with overall survival, and four genes were identified as independent prognostic factors. A prognostic risk score was then proposed, which displayed moderate accuracy in the training and self-validation cohorts. Furthermore, patients in two independent microarray cohorts were successfully stratified into high- and low-risk prognostic groups. Thus, we constructed a reliable prognostic model for pancreatic cancer, which should be beneficial for clinical therapeutic decision-making.
Keywords: TCGA; WGCNA; pancreatic cancer; prognostic prediction model; robust rank aggregation.