Abstract
Autoantibodies against tumor-associated antigens (TAAs) are attractive non-invasive biomarkers for detection of cancer due to their inherently stable in serum. Serum autoantibodies against 9 TAAs from gastric cancer (GC) patients and healthy controls were measured by enzyme-linked immunosorbent assay (ELISA). A logistic regression model predicting the risk of being diagnosed with GC in the training cohort (n = 558) was generated and then validated in an independent cohort (n = 372). Area under the receiver operating characteristic curve (AUC) was used to assess the diagnostic performance. Finally, an optimal prediction model with 6 TAAs (p62, c-Myc, NPM1, 14-3-3ξ, MDM2 and p16) showed a great diagnostic performance of GC with AUC of 0.841 in the training cohort and 0.856 in the validation cohort. The proportion of subjects being correctly defined were 78.49% in the training cohort and 81.99% in the validation cohort. This prediction model could also differentiate early-stage (stage I-II) GC patients from healthy controls with sensitivity/specificity of 76.60%/72.34% and 80.56%/79.17% in the training and validation cohort, respectively, and the overall sensitivity/specificity for early-stage GC were 78.92%/74.70% when being combined with two cohorts. This prediction model presented no significant difference for the diagnostic accuracy between early-stage and late-stage (stage III - IV) GC patients. The model with 6 TAAs showed a high diagnostic performance for GC detection, particularly for early-stage GC. This study further supported the hypothesis that a customized array of multiple TAAs was able to enhance autoantibody detection in the immunodiagnosis of GC.