Pancreatic Cancer Risk Prediction Model Using a Multi-biomarker Panel

Abstract

The pancreatic ductal adenocarcinoma (PDAC) has dismal survival rate due to late detection. Thus, many researches have been tried to discover diagnostic biomarkers for early detection of PDAC. Previously, we developed the triple marker panel, including leucine-rich alpha-2 glycoprotein (LRG1), transthyretic (TTR), and CA 19-9, with sensitivity of 82.5%; specificity of 92.1% 1. Now, in this study, using this panel, we were to discover a risk prediction model for pancreatic cancer for early diagnosis with the enzyme-linked immunosorbent assay (ELISA). There were 744 samples assessed with ELISA, including PDACs (n=396) and normal samples (n=348). We proposed a risk prediction model with machine-learning method, logistic regression (LR), and compared it with support vector machine (SVM) and random forest (RF). To commercialize this model, we searched two optimal thresholds to distinguish three risk groups (high, intermediate, and low) that reliably satisfy four measurements, negative predictive value (NPV), positive predictive value (PPV), sensitivity (SEN), and specificity (SPE), simultaneously greater than 0.95%. The Pearson correlation between the triple marker panel examined with ELISA and the individual marker panel examined with ELISA 1 was 0.884. The risk prediction model distinguished pancreatic cancer from normal individuals with AUC 0.935. The thresholds in between low, intermediate and high groups were 0.11 and 0.77, that satisfied NPV 95.15%, PPV 97.55%, SEN 97.55%, and SPE 95.15%.We first validated reproducibility of the performance of the triple marker panel in this study. And our risk prediction model for pancreatic cancer achieved high accuracy prediction, which can be easily used in the clinic.