Objective: To establish a predictive model for pancreatic cancer by using serum protein footprint map.
Methods: Surface-enhanced laser desorption/ionization (SELDI) ProteinChip technology was applied to screen abnormally expressed proteins in 92 pancreatic cancer patients, 15 patients with benign pancreatic diseases, 18 patients with other diseases of digestive tract, and 96 healthy volunteers. 48 patients with pancreatic diseases and 48 healthy volunteers selected randomly constituted the training group, and 43 pancreatic cancer patients,48 healthy volunteers, 15 patients with pancreatic diseases, and 18 patients with malignant diseases of digestive tract selected randomly constituted the testing group. A predictive model was founded by bioinformation methods using the data of the training group. The data of the testing group was imported to this model to verify by blind method. The sensitivity and specificity of both diagnosis models and CA19-9, were analyzed to evaluate the clinical application value of the model in diagnosis of pancreatic cancer.
Results: Using IMAC3-Cu chip, 12 peaks of differentially expressed proteins were discovered. Six of those had significant value in diagnosis of pancreatic cancer. With the principle of decision tree, a model was founded to diagnose pancreatic cancer, which was composed of 4 decisive nodes and 5 terminal nodes. With the blind test, the sensitivity of the model was 90.7% (39/43) and the specificity was 89.6% (43/ 48) in differentiating cancer from normal status. The modal was better than CA19-9. Serial tests could raise the specificity to 97.9% (47/48), while parallel test could raise the sensitivity to 95.3% (41/43).
Conclusion: The predictive model established by serum protein footprint map is a quick, easy, convenient and accurate method to diagnosis pancreatic cancer and screen new tumor markers.