Fig. 1

ROC curve analysis. ROC curve 1, which was constructed based on the model 1 that included age, sex, education, and the four SNPs (NR5A2 rs2816938, MYC rs10094872, PDX1 rs9581943 and a chromosome 13q22 SNPs, rs4885093), had an AUC = 0.7115. ROC curve 2, which was constructed based on the model 2 that included age, sex, education, and the seven environmental factors (cigarette smoking, oral hygiene score, vegetable consumption, fruit consumption, allergy, diabetes/glucose intolerance, and BMI) had an AUC = 0.7818, which was significantly better than the AUC of the ROC curve 1 (p < 0.0001). ROC curve 3, which was constructed based on the model 3 that combined the 7 environmental factors and the four SNPs plus age, sex, and education, improved the AUC to 0.7924, which was marginally better the AUC of ROC curve 2 (p = 0.05). ROC curve 4, which included age, sex, education, the 7 environmental factors, the four SNPs and the two significant gene-environment interactions (poor oral hygiene x NR5A2 rs2816938 and BMI x PDX1 rs9581943) improved the AUC further to 0.7982, which was significantly better than the AUC of ROC curve 2 (p = 0.006)