A Translational Model to Improve Early Detection of Epithelial Ovarian Cancers

Abstract

Neural network analyses of circulating miRNAs have shown potential as non-invasive screening tests for ovarian cancer. A clinically useful test would detect occult disease when complete cytoreduction is most feasible. Here we used murine xenografts to sensitize a neural network model to detect low volume disease and applied the model to sera from 75 early-stage ovarian cancer cases age-matched to 200 benign adnexal masses or healthy controls. The 14-miRNA model efficiently discriminated tumor bearing animals from controls with 100% sensitivity down to tumor inoculums of 50,000 cells. Among early-stage patient samples, the model performed well with 73% sensitivity at 91% specificity. Applied to a population with 1% disease prevalence, we hypothesize the model would detect most early-stage ovarian cancers while maintaining a negative predictive value of 99.97% (95% CI 99.95%-99.98%). Overall, this supports the concept that miRNAs may be useful as screening markers for early-stage disease.