Introduction : The diagnostic value of a single tumor marker is not high in pancreatic cancer. To overcome this limitation, combination of multiple biomarkers and unique analysis algorithm are used. The purpose of this study was to develop diagnostic biomarker panel algorithms and validated their performance diagnosing pancreatic ductal adenocarcinoma

Methods : From October 2002 to March 2014, 180 PDAC and 573 normal control serum samples were used. Candidate markers consisted of 35 markers expressed in 6 cancers of Korean people, hepatocellular carcinoma, breast cancer, lung cancer, gastric cancer, colon cancer, and prostate cancer. Two-thirds of the samples were taken as a training set, validation was performed with the remaining one-third set. Five linear and nonlinear classification algorithms are used to select the optimal model

Results : There were difference in 11 marker of 35 marker (ApoA1, CA125, CA19-9, B2M, CRP, CYFRA21.1, LRG1, CEA, ApoA2, TTR and D.Dimer). Using 11 markers, 2047 combinations were made and all combinations were applied to 5 algorithms. The optimal multiple biomarker combination was consist of 6 marker ApoA1, CA125, CA19-9, CEA, ApoA2, and TTR with random forest classification algorithms method. The AUC, sensitivity and specificity were 0.992, 96% and 95% and in training set, and those were 0.992, 93% and 96% in validation set.

Conclusions : Combination of multiple biomarkers for the diagnosis of pancreatic cancer from Korean cancer panel were established. As a result, the early diagnosis rate of pancreatic cancer can be increased, and improved survival rates can be expected.