Aims Tumour markers including carcinoembryonic antigen (CEA) or carbohydrate antigen 19-9 (CA19-9) are frequently determined at the time of diagnosis in patients with pancreatic cancer. Several studies indicate a prognostic relevance of these markers in pancreatic cancer, but space for improvement with regard to the predictive accuracy and ability is given. In this work, the main focus is on mathematical combinations of these two tumour markers in order to validate an improvement of prognostic test results in terms of sensitivity and specificity.

Methods This retrospective study includes 393 patients with pancreatic cancer, who were treated between the years 2005 and 2012 at the Division of Oncology, Medical University of Graz, Austria. The goal of this study was to explore whether an appropriate combination of two tumour markers leads to a statistically significant improvement of the prognostic prediction.

Results Receiver operating characteristic curves comparison analyses with the classification variable cancer-specific survival showed that the mathematical product of two tumour markers (TM_product= (CEA×CA19-9); area under the curve (AUC)=0.727; 95% CI 0.680 to 0.770) is significantly better than CEA alone (AUC=0.644; 95% CI 0.594 to 0.691; p=0.003) but not significant compared with CA19-9 (AUC=0.710; 95% CI 0.662 to 0.754; p=0.1215). A linear combination of CEA and CA19-9 (TM_linear=(85×CEA+CA19-9); AUC=0.748; 95% CI 0.702 to 0.790) is significantly better than CEA (p<0.0001) as well as CA19-9 alone (p=0.0304).

Conclusions Mathematical combinations of pretherapeutic tumour markers CEA and CA19-9 are feasible and can significantly improve the prognostic prediction in patients with pancreatic cancer.