DEVELOPMENT AND VALIDATION OF A MACHINE-LEARNING-DERIVED RNASEQ PROGNOSTIC SIGNATURE IN ENDOMETRIAL CANCER

Introduction/Background* Because of inter-tumor heterogeneity of endometrial carcinoma (EC), prognostication remains challenging. We aimed to develop a RNAseq signature to stratify EC patient prognosis beyond molecular subtyping. Methodology A prognostic signature was identified using a LASSO-penalized Cox regression model on TCGA (N=543 patients). A polyA-RNAseq-based method was developed for validation of the signature in a cohort of stage I-IV EC patients treated in two Paris Hospitals between 2010 and 2017. Model performances were evaluated using time-dependent ROC curves (prediction of disease-specific-survival (DSS)). The additional value of the RNAseq signature was evaluated using uni/multivariable Cox models (hazard ratio (HR) with [95% confidence interval]) and Kaplan-Meier analysis. Result(s)* Among 209 patients included in the validation cohort (median follow-up 55 months IQR [41-69]), 61 (30%), 10 (5%), 52 (25%), and 82 (40%), had mismatch repair-deficient, POLE-mutated, TP53-mutated tumors, and tumors with no specific molecular profile, respectively. The 38-genes signature accurately predicted DSS (AUC=80%). Using a composite classifier accounting for the RNAseq signature and the TP53-mutated group, three groups were identified: good prognosis tumors based on RNAseq signature and without TP53 mutation, characterized by excellent outcome (N=103 patients, 5-years DSS rates of 99%) (reference), poor prognosis tumors based on RNAseq signature and without TP53 mutation (N=49 patients, 5-years DSS rates of 81%; HR: 5.86 [1.16; 29.7]), and TP53-mutated tumors whatever the RNAseq signature (N=52 patients, 5-years DSS rates of 52%; HR: 11.14 [2.40; 51.7]) (HR adjusted on FIGO stage). In 81 (38%) patients with adverse features (2020 ESGO/ESTRO/ESP guidelines: non-endometrioid histology or stage III-IVA or TP53-mutated tumors), TP53-mutated molecular group was not significantly associated with poor prognosis (p=0.18). A The composite classifier identified three classes within this subgroup: RNAseq-good prognosis (N=24), non-TP53/RNAseq-poor prognosis (N=16), and TP53-mutated tumors (N=41), with 5-years DSS rates of 100%, 59%, and 71%, respectively (p=0.015). Transcriptome analyses suggested the underlying involvement of immune deprivation and wound healing processes in tumors with poor prognosis. Conclusion* We demonstrate that RNAseq characterization can refine prognostication in EC beyond molecular subgroups and main prognostic features, and warrants validation for potential RNAseq-based adjuvant therapeutic strategies in EC.