Novel Model for Comprehensive Assessment of Robust Prognostic Gene Signature in Ovarian Cancer Across Different Independent Datasets

Different analytical methods or models can often find completely different prognostic biomarkers for the same cancer. In the study of prognostic molecular biomarkers of ovarian cancer (OvCa), different studies have reported a variety of prognostic gene signatures. In the current study, based on geometric concepts, the linearity-clustering phase diagram with integrated P-value (LCP) method was used to comprehensively consider three indicators that are commonly employed to estimate the quality of a prognostic gene signature model. The three indicators, namely, concordance index, area under the curve, and level of the hazard ratio were determined via calculation of the prognostic index of various gene signatures from different datasets. As evaluation objects, we selected 13 gene signature models (Cox regression model) and 16 OvCa genomic datasets (including gene expression information and follow-up data) from published studies. The results of LCP showed that three models were universal and better than other models. In addition, combining the three models into one model showed the best performance in all datasets by LCP calculation. The combination gene signature model provides a more reliable model and could be validated in various datasets of OvCa. Thus, our method and findings can provide more accurate prognostic biomarkers and effective reference for the precise clinical treatment of OvCa.