Factors Requiring Adjustment in the Interpretation of Serum Carcinoembryonic Antigen: A Cross-Sectional Study of 18,131 Healthy Nonsmokers

Abstract

Serum carcinoembryonic antigen (CEA) is a well-known tumor marker for colorectal adenocarcinoma. However, CEA levels can be influenced by various nonmalignant conditions. A retrospective, cross-sectional study was performed including 18,131 healthy nonsmokers who underwent health check-ups with evaluation of the serum CEA level. In the training set, multivariate analysis revealed that the log-transformed CEA level had positive relationships with age (regression coefficient (r) = 0.005, P<0.001), white blood cell (WBC) count (r=0.007, P=0.016), hemoglobin (HB, r=0.016, P<0.001), aspartate aminotransferase (AST, r=0.002, P=0.005), creatinine (r=0.076, P=0.038), and glycosylated hemoglobin (HbA1c, r=0.052, P<0.001); body mass index (BMI, r=−0.007, P<0.001) showed a negative correlation. The results for age, BMI, WBC count, HB, AST, and HbA1c were validated in the test set. We were able to construct the following model to predict the log-transformed CEA level: log (CEA + 0.51) = −0.204 − 0.051 (gender) + 0.005 (age) − 0.006 (BMI) + 0.008 (WBC count) + 0.016 (HB) + 0.002 (AST) + 0.062 (creatinine) + 0.054 (HbA1c). For colorectal cancer prediction, the model with the observed CEA and adjusted CEA levels had significantly high predictive power (AUC 0.756, P<0.001) than the model only including the observed CEA level (AUC 0.693, P<0.001). Factors influencing serum CEA levels should be adjusted before clinical interpretation to increase the predictive value of CEA.