Diagnosis of epithelial ovarian cancer using a combined protein biomarker panel

Abstract

Background: An early detection tool for EOC was constructed from analysis of biomarker expression data from serum collected during the UKCTOCS. Methods: This study included 49 EOC cases (19 Type I and 30 Type II) and 31 controls, representing 482 serial samples spanning seven years pre-diagnosis. A logit model was trained by analysis of dysregulation of expression data of four putative biomarkers, (CA125, phosphatidylcholine-sterol acyltransferase, vitamin K-dependent protein Z and C-reactive protein); by scoring the specificity associated with dysregulation from the baseline expression for each individual. Results: The model is discriminatory, passes k-fold and leave-one-out cross-validations and was further validated in a Type I EOC set. Samples were analysed as a simulated annual screening programme, the algorithm diagnosed cases with >30% PPV 1–2 years pre-diagnosis. For Type II cases (~80% were HGS) the algorithm classified 64% at 1 year and 28% at 2 years tDx as severe. Conclusions: The panel has the potential to diagnose EOC one-two years earlier than current diagnosis. This analysis provides a tangible worked example demonstrating the potential for development as a screening tool and scrutiny of its properties. Limits on interpretation imposed by the number of samples available are discussed.