Introduction Ovarian cancer is the most fatal of all female reproductive cancers, thus a new reliable and accurate screening test for ovarian cancer is urgently demanded. We established a think-outside-the-box screening method that combines cancer-related tumor markers and comprehensive glycan alterations in serum glycoproteins, which represent a physical state (CSGSA: comprehensive serum glycopeptide spectra analysis). We aimed to verify the diagnostic capability of CSGSA, a blood test to identify early-stage epithelial ovarian cancer (EOC) in this study.

Methods We obtained sera of 564 EOC patients (55.8 ± 12.2 years) and 1,154 non-EOC controls (54.1 ± 12.0 years) from 13 facilities. Expression patterns of 1,712 glycopeptides detected by liquid chromatography mass spectrometry (LC-MS) and cancer-related tumor markers were analyzed by convolutional neural network (CNN) to discriminate an early-stage EOC.

Results CSGSA CNN model discriminated early-stage EOC (Stage I) from non-EOC controls with ROC-AUC 0.929 (95% CI: 0.919–0.940), which exceeded those of current tumor markers, CA125 (0.840, 95% CI: 0.811–0.870) and HE4 (0.718, 95% CI: 0.675–0.760). Positive predictive value (PPV) correlated by the prevalence became 7.1% where EOC sensitivity was 51.7%.

Conclusion/Implications We confirmed that the CSGSA discriminated early-stage EOC with high sensitivity and specificity. It is expected to identify early-stage EOC in asymptomatic women before EOC develops to advanced stage.