266 Models to predict response to hormonal therapy in patients with recurrent or advanced endometrial cancer

• Published in: International journal of gynecological cancer Vol. 34; no. Suppl 1; pp. A168 - A169
• Main Authors: Jin, Xiaoman, Silvertand, Daphne, Werner, Henrica M.j., Pijnenborg, Johanna M.a., Lalisang, Roy I., Bulten, Johan, Eriksson, Ane G.z., Lindemann, Kristina, Van Beekhuizen, Heleen J., Trum, Hans, Witteveen, Petronella. O., Galaal, Khadra, Ginkel, Alexandra Van, Weinberger, Vit, Sweegers, Sanne, Krakstad, Camilla, Weelden, Willem Jan Van, Fijten, Rianne, Romano, Andrea
• Format: Journal Article
• Language: English
• Published: Oxford BMJ Publishing Group Ltd 10.03.2024; BMJ Publishing Group LTD
• Subjects: Biomarkers; Drugs; Endometrial cancer; Poster and E-Posters; Response rates
• ISSN: 1048-891X; 1525-1438
• DOI: 10.1136/ijgc-2024-ESGO.318

Introduction/BackgroundThere is no clear guidance for systemic treatment in patients with advanced/recurrent endometrial cancer (EC). Hormonal therapies can be considered in a palliative setting, yet there is a lack of biomarkers to predict a therapeutic response to the drug. This study aimed to identify effective biomarkers from tumor transcriptomics and develop artificial intelligence (AI) models which can predict the therapeutic response of patients to hormonal drugs.MethodologyThe PROMOTE study population was previously described (van Weelden et al, AJOG 2021) and included patients with advanced stage/recurrent EC treated with hormonal drugs. Tumor samples from a total of 61 patients (out of the full cohort of 102 eligible patients) with sufficient isolated RNA were subjected to RNA-seq (Illumina). Patients were grouped according to their response to Clinical Benefit Rate (CBR: complete response, partial response and stable disease) and Response Rate (RR: complete response, partial response) were computed. Univariate analysis based on DESeq2 method and multivariate analysis based on principle component analysis and recursive feature elimination were applied using R Studio.ResultsA total of 97 differentially expressed genes were identified for CBR and 16 (10 upregulated and 6 downregulated) showed a fold-change >4; 103 differentially expressed genes were identified for RR with 24 (16 upregulated and 8 downregulated) showing a fold-change higher than 4. Interestingly, genes involved in the steroid hormone metabolism like HSD17B3, AKR1C2 were differentially expressed in relation to response to hormonal drugs. Prediction models were developed either using transcriptomic data only or after combining transcriptomics with clinical features (age, stage, grade etc.). CBR and RR could be predicted with a good accuracy on the training and test data sets.ConclusionBased on RNA seq data on pretreatment tumor biopsies the response rate to hormonal drugs can be predicted with a good accuracy.DisclosuresNone