Background: Recent advances in extracellular vesicle (EV)-associated biomarker discovery, particularly miRNA profiling integrated with advanced machine learning algorithms, have demonstrated promising potential for developing robust multivariate classification models that more accurately detect tumour-specific molecular signatures in early-stage disease.
Aim: To test the hypothesis that a machine learning-optimised multivariate index assay combining EV subpopulation miRNA analysis with CA125 plasma concentrations will meet established screening criteria (>75% sensitivity, >99% specificity) for ovarian cancer detection in asymptomatic women, demonstrating significantly improved accuracy over current modalities.
Methods: Peripheral blood samples were collected from asymptomatic, non-smoking, average-risk women (controls) and women undergoing clinical investigations for suspected gynaecological disorders. Plasma samples were stratified by histologically confirmed ovarian cancer (Stages I-IV, n=123), benign conditions (n=154), and healthy controls (n=220). Samples were randomised into training (75%) and test (25%) sets. EV-associated miRNA was isolated using EXO-NET™ and Promega Maxwell HT kits on an automated KingFisher Apex system. miRNA targets were quantified by RT-qPCR. Plasma CA125 concentrations were determined using commercially available ELISA. The optimised MIA was validated in an independent cohort (n=125).
Results: For the independent test set, ROC curve analysis reported and area under the curve (AUC) of 0.971 and an overall of sensitivity 77% at >99.6% specificity. Of note, the MIA correctly identified all case of early (Stages I/II) disease and outperformed CA125 alone (61% sensitivity overall; 44% for Stage I).
Conclusions: This study supports the hypothesis that an automated, machine learning-optimised MIA utilising EV-associated miRNA and plasma CA125 can achieve classification performance required for early ovarian cancer detection. Screening tests for asymptomatic, average-risk women require high specificity (>99.6%) to minimise false positives. The achievement of 100% sensitivity for early-stage ovarian cancer detection, combined with >99% specificity, represents significant progress in developing clinically meaningful screening. The potential survival benefit, increasing 5-year survival from approximately 49% to over 90% through early detection, highlights the transformative clinical impact. These findings establish EV-based multivariate analysis as a promising strategy for effective ovarian cancer screening in average-risk populations.