Validation of miRNA signatures for ovarian cancer earlier detection in the pre-diagnosis setting using machine learning approaches.

Introduction: Effective strategies for early detection of epithelial ovarian cancer are lacking. We evaluated whether a panel of 14 previously established circulating microRNAs could discriminate between cases diagnosed <2 years after serum collection and those diagnosed 2-7 years after serum collection. miRNA sequencing data from subsequent ovarian cancer cases were obtained as part of the ongoing multi-cancer JanusRNA project, utilizing pre-diagnostic serum samples from the Janus Serum Bank and linked to the Cancer Registry of Norway for cancer outcomes. Methods: We included a total of 80 ovarian cancer cases contributing 80 serum samples and compared 40 serum samples from cases with samples collected <2 years prior to diagnosis with 40 serum samples from cases with sample collection ≥2 to 7 years. We employed the extreme gradient boosting (XGBoost) algorithm to train a binary classification model using 70% of the available data, while the model was tested on the remaining 30% of the dataset. Results: The performance of the model was evaluated using repeated holdout validation. The previously established set of miRNAs achieved a median area under the receiver operating characteristic curve (AUC) of 0.771 in the test sets. Four out of 14 miRNAs (hsa-miR-200a-3p, hsa-miR-1246, hsa-miR-203a-3p, hsa-miR-23b-3p) exhibited higher expression levels closer to diagnosis, consistent with the previously reported upregulation in cancer cases, with statistical significance observed only for hsa-miR-200a-3p (beta=0.14; p=0.04). Discussion: The discrimination potential of the selected models provides evidence of the robustness of the miRNA signature for ovarian cancer.